Professional Documents
Culture Documents
(Lecture Notes in Economics and Mathematical Systems) Sebastian Rausch - Macroeconomic Consequences of Demographic Change - Modeling Issues and Applications-Springer (2009)
(Lecture Notes in Economics and Mathematical Systems) Sebastian Rausch - Macroeconomic Consequences of Demographic Change - Modeling Issues and Applications-Springer (2009)
(Lecture Notes in Economics and Mathematical Systems) Sebastian Rausch - Macroeconomic Consequences of Demographic Change - Modeling Issues and Applications-Springer (2009)
Managing Editors:
Prof. Dr. G. Fandel
Fachbereich Wirtschaftswissenschaften
Fernuniversität Hagen
Feithstr. 140/AVZ II, 58084 Hagen, Germany
Prof. Dr. W. Trockel
Institut für Mathematische Wirtschaftsforschung (IMW)
Universität Bielefeld
Universitätsstr. 25, 33615 Bielefeld, Germany
Editorial Board:
A. Basile, H. Dawid, K. Inderfurth, W. Kürsten
Sebastian Rausch
Macroeconomic Consequences
of Demographic Change
Modeling Issues and Applications
123
Dr. Sebastian Rausch
Postdoctoral Associate
Joint Program on the Science and Policy
of Global Change
Massachusetts Institute of Technology
77 Massachusetts Ave, Bldg E19-439a
Cambridge, MA 02139-4307
USA
ISSN 0075-8442
ISBN 978-3-642-00145-1 e-ISBN 978-3-642-00146-8
DOI 1 0.1007/978-3-642-00146-8
Springer Dordrecht Heidelberg London New York
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Macroeconomic Consequences of Demographic Change: Focus
and New Aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.1 Global Demographic Change and International Trade . . . . . . 7
1.2.2 Quantifying the Sectoral and Distributional Effects of
Population Aging in Germany . . . . . . . . . . . . . . . . . . . . . . . . . 8
4.5.1
Age-specific Preferences: How Important Are Structural
Changes in Final Consumption Demand ? . . . . . . . . . . . . . . . . 114
4.5.2 Piecemeal Sensitivity Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 117
4.6 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
Appendix 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
Appendix 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
List of Figures
2.1 Solution to the initial representative agent model (Step 0 and Step 1) 19
2.2 Evaluating household demands at new prices (Step 2) . . . . . . . . . . . . . 20
2.3 Recalibration of preferences (Step 3) . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.4 Iterative adjustment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.5 Comparison of SR and tâtonnement fields . . . . . . . . . . . . . . . . . . . . . . . 26
2.6 Global convergence behavior for different configurations of θ and σ . 27
2.7 Solving OLG by Ramsey: steps in the decomposition algorithm . . . . . 31
2.8 Approximating OLG by Ramsey: the issue of terminal generations . . 34
2.9 Solving OLG by Ramsey: sequence of investment time paths . . . . . . . 38
2.10 Solving OLG by Ramsey: sequence of welfare changes by generation 39
2.11 Solving OLG by Ramsey: approximation error ek . . . . . . . . . . . . . . . . . 40
The world is in the midst of a major demographic transition. Not only is the popu-
lation growth slowing, but the age structure of the population is changing, with the
share of the young falling and that of the elderly rising. The sources of population
aging lie in two demographic phenomena: rising life expectancy and declining fer-
tility. Population aging is most advanced in the industrialized countries but, with
a lag, demographic trends in many developing economies will follow (United Na-
tions, 2006). Because many developing countries are experiencing faster fertility
transitions, in the future they will experience even faster population aging than the
currently developed countries (Bloom and Williamson (1998) and Weil (2006)).
In the coming decades, this demographic transition will prove to be a key factor in
the development of societies. From a macroeconomic perspective, the consequences
of population aging are substantial and manifold. For example, a lower growth in
the working age population causes aggregate labor supply to decline and reduces
the amount of investment required to supply new workers with capital. As a result,
capital becomes abundant and labor becomes scarce with ensuing changes in factor
prices and interest rates (Krueger and Ludwig, 2007). Börsch-Supan (2003) argues
it is unlikely that the decline in the labor force will be offset by higher capital in-
tensity to prevent adverse effects on domestic production. This will have significant
impacts on labor markets (Börsch-Supan, 2003) as well as capital markets (Kot-
likoff, Smetters and Walliser, 2001) influencing savings and investment behavior
and ultimately the growth rate of the economy (Batini, Callen and McKibbin, 2006).
Demographic change also interacts with public policies through various channels.
Most obviously, rising old age dependency ratios jeopardize the sustainability of cur-
rent social security systems (Abel, 2003). Burtless (2006) calculates that the effect
of population aging would significantly raise tax rates required to finance increased
government spending. With respect to goods markets, populating aging is bound
to change the composition of aggregate demand thereby changing relative prices
(Lührmann, 2005).
While this list of consequences is not by any means exhaustive, an abundant body
of literature has scrutinized many of the possible effects of demographic change;
see, e.g., Börsch-Supan (2004) for an overview. In order to quantify these effects
1 For surveys of econometric studies that deal with demographic change see, e.g., Bryant and
McKibbin (1998, 2003).
2 See Kotlikoff (2000) for an overview. The basic model in which households live for two periods
equilibrium models, the OLG framework is a natural candidate for introducing het-
erogeneous agents into the economic analysis. By construction, OLG models feature
heterogeneity of age among agents, and adding intra-cohort heterogeneity to these
models involves a—more or less—trivial indexing of agents within each generation.
Advancing numerical OLG models in this direction is clearly desirable because it
strengthens the micro-foundation of this type of analysis, and allows to analyze in
more detail the distributional implications of economic policy. Not surprisingly, this
is an active area of research. In a considerable number of papers, e.g., Conesa and
Krueger (1999), Kotlikoff, Smetters and Walliser (1999), Krusell and Smith (1998)
Huggett and Ventura (1999), Jensen and Rutherford (2002), Krueger and Ludwig
(2007) and Storesletten, Telmer and Yaron (2007).
1.1 Methodology
Large-scale OLG models have been fruitfully used to analyze a wealth of economic
issues, as for example tax reform, investment incentives, government spending and
debt, social security reform, monetary policy, endogenous growth, human capital ac-
cumulation, and demographic change. However, the scope in many applications and
the required degree of complexity are often limited by the computational power of
available numerical solution techniques. Quantitative simulation models with over-
lapping generations naturally involve a large number of variables and equations that
describe the equilibrium behavior of economic agents and the aggregate economy. In
particular, models that exhibit a rich household side including a variety of household-
specific effects, a large number of heterogeneous households, and realistic lifetimes
of agents, typically require “customized” solution methods that may both be costly
to implement and difficult to validate. Under such circumstances, the decision to add
specific features to the model may not be driven by economic reasoning alone but
also by the need to take into account the computational feasibility of the numerical
problem.
Chapter 2 of this research monograph, written with Thomas F. Rutherford, aims
at remedying this problem by developing a new decomposition algorithm that can
be used to efficiently solve large-scale OLG models. The algorithm can be imple-
mented using “off the shelf numerical tools” which are routinely available through
GAMS.4 The presented approach is primarily appropriate for computing equilibria
in models in which the number of agents is so large that simultaneous solution meth-
ods operating directly on the equilibrium system of equations are infeasible due to
of the Ramsey model. The analyses in this research monograph, however, focus on OLG economies
that are dynamically efficient.
4 GAMS (General Algebraic Modelling System) is a computer language originally developed to
assist economists at the World Bank in the quantitative analysis of economic policy questions
(Meeraus (1983) and Brooke, Kendrick and Meeraus (1988)). Currently, two large-scale solvers
are available through GAMS/MCP : MILES employs a modified Newton algorithm and PATH is
based on a path-following procedure. For details and more references see Rutherford (1995b).
4 1 Introduction
the high dimensionality related to income and household-specific effects. The Se-
quential Recalibration (SR) algorithm is based on the solution of a sequence of non-
linear complementarity problems. A characteristic of many economic models is that
they can be cast as a complementary problem. Rutherford (1995b) and Mathiesen
(1985) have shown that a complementary-based approach is convenient, robust, and
efficient.5
The main idea of the proposed algorithm is to solve a market economy with many
households through the computation of equilibria for a sequence of representative
agent economies. The dimensionality of the numerical problem is significantly re-
duced by an appropriate decomposition of the system of equations that defines the
dynamic general equilibrium of the OLG economy. To achieve this, the OLG econ-
omy is transformed into a lower-dimensional Ramsey growth problem by replacing
the entire system of overlapping generations with a single infinitely lived represen-
tative agent. Apart from this modification, both models are identical. The first step
in each iteration is to compute a vector of general equilibrium prices given the eco-
nomic policy (or shock) under consideration. In a second step, the candidate prices
are used to evaluate the demand functions of the actual overlapping generations
households. This amounts to solving a partial equilibrium relaxation of the original
OLG economy that suppresses all general equilibrium interactions. The final step in
each iteration uses households’ quantity choices to recalibrate, i.e. to choose a set of
preferences for the artificial Ramsey agent such that, given candidate prices, aggre-
gated quantity choices by OLG households are consistent with the equilibrium be-
havior of the representative agent. The SR algorithm finds the equilibrium allocation
of the underlying OLG economy if the sequence of prices and quantities obtained
from the Ramsey growth model and the partial equilibrium relaxation converge to
the true equilibrium prices and quantities.
The decomposition algorithm provides improvements in both efficiency and ro-
bustness as compared with simultaneous solution methods. To demonstrate this
point, a full-blown Auerbach-Kotlikoff OLG model with production activities, cap-
ital accumulation, endogenous labor supply, and a government sector is augmented
by a large number of heterogeneous households. Evaluating the algorithm’s perfor-
mance compared with an “integrated” complementarity-based approach as put for-
ward by Rasmussen and Rutherford (2004), shows that the decomposition algorithm
is able to solve high-dimensional OLG models that are infeasible for simultaneous
solution methods. Overall, the quality of approximation is found to be excellent and
decreases marginally in the degree of intra-cohort heterogeneity.
To characterize limitations of the SR algorithm, local convergence theory for a
simple exchange economy due to Scarf (1960) is developed, and global conditions
for convergence failure are determined by means of numerical analyses. The condi-
tions for local stability of the implied adjustment process reduce to those of a Wal-
rasian price tâtonnement process. Thus, the proposed algorithm belongs to a large
5 The complementarity format embodies weak inequalities and complementary slackness, relevant
features for models that contain bounds on specific variables, e.g. activity levels which cannot
a priori be assumed to operate at positive intensity. Such features are not easily handled with
alternative solution methods.
1.2 Macroeconomic Consequences of Demographic Change: Focus and New Aspects 5
class of algorithms used in computational economics that are robust, efficient, and
yet fail to provide global convergence. In general, however, the appropriateness of
the proposed algorithm depends on the characteristics of the underlying economic
model. A central result is that the algorithm may be inappropriate for applications in
which there are significant income effects. This deficiency is a consequence of the
simplifying nature of the adjustment process which omits both income constraints
and global characteristics of the individual utility functions.
Notwithstanding these limitations, the decomposition algorithm can be beneficial
for a wide range of economic applications. Due to its decomposition strategy, which
operates on the household side of the economy, it is particularly tailored to models
with many heterogenous households. As such, the SR algorithm can be instrumen-
tal in developing (and solving) large-scale OLG models to analyze in considerable
detail both the inter- and intragenerational distributional implications of economic
policy. In general, however, the decomposition method can also be advantageous for
modeling tasks that need to economize on the dimensionality of the numerical prob-
lem. For instance, multi-sectoral and multi-regional OLG models typically involve
a large number of equations and variables that often render simultaneous solution
methods infeasible. By transforming the OLG model into a lower-dimensional Ram-
sey growth problem, and applying the decomposition approach, the SR algorithm
can create sufficient degrees of freedom in the model architecture to accommodate
such complex research designs.
The proposed algorithm also contributes to the growing literature on the inte-
gration of “micro-macro” models that combine the strengths of both computable
general equilibrium and microsimulation models. The SR algorithm does not only
overcome a principal weakness that is characteristic of many approaches in this
strand of literature (see, e.g., Bourguignon, Robilliard and Robinson (2005) and
Bourguignon and Spadaro (2006)), namely to neglect feedback effects from the mi-
cro to the macro level. Applied to the OLG context, it is also the first computational
device to integrate both approaches in a dynamic framework.
Chapter 3 and Chapter 4 are based on the methodology developed in Chapter 2.
The OLG models developed in both chapters exploit the advantages of the decom-
position algorithm to study specific aspects of the macroeconomic implications of
population aging that have received little attention in the literature. Hence, although
each chapter focuses on economic analysis rather than on computational issues, they
both demonstrate the flexibility, scope, and power of the SR algorithm.
In many industrialized countries populations are aging, and the demographic tran-
sition will prove to be one of the key factors in shaping the development of soci-
eties in the coming decades. This process is driven by falling mortality rates and
a decline in birth rates, which reduces population growth rates and increases the
6 1 Introduction
share of older people in the economy. While an extensive economic literature has
scrutinized many aspects of the possible economic effects of population aging, see
Börsch-Supan (2004) for an overview, this book relates to the strand of literature
that uses the models and techniques pioneered by Auerbach and Kotlikoff (1987).
From a macroeconomic perspective, the most striking consequence is the change
in the relative abundance of factors of production. As the demographic transition
reduces population growth and decreases the share of young people in the economy,
the aggregate labor force is projected to decline. At the same time, a larger share
of older people—who tend to hold a larger amount of financial wealth—increases
the amount of capital in the economy. Both effects imply that capital in an aging
society becomes abundant relative to labor. Another way of thinking about this is to
imagine that in the light of a shrinking labor force the accumulated capital stock does
not depreciate fast enough to maintain a constant equipment of physical capital per
worker. In other words, demographic change is bound to increase the capital-labor
ratio—often referred to as the process of “capital deepening” (Kotlikoff et al., 2001).
Everything else equal, factor prices will change to reflect the varying scarcities of
capital and labor: the capital rental rate is projected to decline, while wage rates are
bound to increase.
While this adjustment mechanism to demographic change is extensively docu-
mented in the literature, and is at the heart of most analyses, the majority of stud-
ies pays special attention to the impact of aging on the viability of social security
systems in closed economies. Important examples in a closed economy are Huang,
Imrohoroglu and Sargent (1997), De Nardi, Imrohoglu and Sargent (1999), Abel
(2003), and Fehr, Halder and Jokisch (2004a), and in an open economy setting At-
tanasio, Kitao and Violante (2006) and Börsch-Supan, Ludwig and Winter (2006).
Although issues related to social security are probably among the most obvious con-
sequences of demographic change, these are by no means the only consequences.
Population aging affects virtually all markets, i.e. goods markets, labor markets and
capital markets.
This book focuses on an analysis of the economic consequences of demographic
change per se and not just the consequences of alternative social security reform sce-
narios. More specifically, it tries to provide both qualitative and quantitative answers
to questions such as the following: What are the aggregate economic consequences
of population aging? How is individual economic behavior affected? How do re-
turns to capital and labor evolve during the demographic transition? How is the sec-
toral composition of output affected? What are the consequences for the intra- and
intergenerational distribution of income, wealth, and welfare? Does demographic
change make households worse-off? As ongoing and projected demographic pro-
cesses are not synchronized across countries or world regions, how is international
trade affected? Is trade liberalization under such demographic circumstances always
beneficial and how are gains from trade are distributed across generations over time
and between different types of households?
1.2 Macroeconomic Consequences of Demographic Change: Focus and New Aspects 7
A salient feature of global demographic change is the fact that the aging of popu-
lations will occur at differing paces and with a differing degree of intensity in the
industrialized countries of the world. Significant aging is already under way in some
economies, for example in Germany, Italy, Japan, while major demographic changes
in the U.S. and China will begin in the second decade of the 21st century, and with
a still longer lag, demographic trends in the developing economies will follow.
While the existing literature consistently identifies the presence of globally un-
synchronized aging patterns as an important driving force of cross-border flows
of capital and labor (see, e.g., Attanasio and Violante (2005), Börsch-Supan et al.
(2006), INGENUE (2001), and Fehr, Jokisch and Kotlikoff (2004b)), it surprisingly
overlooks its implications for international trade in goods. To fill this gap, Chapter 3
develops an augmented version of the canonical Heckscher-Ohlin model to analyze
the economic consequences of trade liberalization in a world that is characterized by
globally unsynchronized aging patterns. Adopting a stylized dynamic two-country
two-sector two-factor model with overlapping generations it shows that unsynchro-
nized demographic patterns emerge as a potential determinant of international trade
flows. The mechanism behind this is simple: since relative factor endowments are
endogenously determined by life cycle saving and labor supply behavior of over-
lapping generations, different extents and timing of demographic processes across
countries imply international differences in relative factor endowments, and hence
induces Heckscher-Ohlin trade patterns.
If demographic differences emerge as a potential determinant of international
trade flows, a natural question arises whether a country or a region can benefit from
diffusing part of its demographic shock by means of liberalizing its trade policy.
Since trade tends to reduce the dispersion in international factor prices, opening up
an economy for trade might mitigate the pressure on factor prices exerted by popu-
lation aging. However, as real wages and interest rates move in opposite directions,
the overall effect on income and welfare of households is ambiguous. Chapter 3
therefore provides a quantitative assessment of the welfare effects from trade liberal-
ization in the presence of globally unsynchronized aging patterns. Given the absence
of any distortion, and given the perfect neoclassical setup of the model that abstracts
from any form of market imperfection, the general result is surprising, standing in
contrast to what would be expected from standard neoclassical trade theory: that
trade liberalization in the presence of globally unsynchronized aging patterns does
not necessarily lead to welfare gains.
In the fast aging country, trade liberalization is only beneficial for older gen-
erations born before and at the beginning of the demographic transition, whereas
future generations stand to incur substantial utility losses. The reason for this is the
evolution of autarky-trade factor price differentials during the demographic transi-
tion. As trade is bound to increase the real return to capital and to decrease the real
wage rate (relative to autarky), older asset-rich households gain while future genera-
tions suffer from losses in labor income. The model also analyzes the distribution of
welfare gains and losses with respect to an intragenerational dimension. In general,
8 1 Introduction
low-skilled and asset-poor households in the fast aging region tend to be relatively
worse off.
Employing Monte-Carlo simulation methods, the results are found to be robust
for a broad range of empirically plausible parameter configurations. It is estimated
that there is a very high probability for households in the fast aging region not to gain
from liberalizing trade. Moreover, trading off gains from trade for older generations
against welfare losses for younger households, it is found that an aggregate welfare
improvement would only materialize in the presence of unrealistically high social
discount rates.
be scarce relative to capital, real wages are bound to increase while real returns
to capital decrease. A priori, it is not clear how this opposite movement of factor
prices is going to affect households’ labor supply and life cycle savings behavior,
and welfare. The quantitative model is used to estimate the welfare effects of the
demographic transition along inter- and intragenerational dimensions, and to assess
how the distribution of labor income and wealth evolves over time.
Over the main projection period from 2003-2050, the demographic transition in
Germany will bring about a 7% decrease in output per capita. This is largely driven
by a significant decrease in aggregate labor supply, about 17% over the same hori-
zon (as the model features a labor-leisure trade-off this number already incorporates
general equilibrium reactions by households), and a substantial decline in invest-
ment rates (by 26% in 2050). Population aging is shown to imply a “capital deepen-
ing” process that is reflected by an increase in the capital-labor ratio of around 2%.
Real factor prices reflect the varying scarcities of factors of production, and the real
capital rental rate (real wage rate) falls (increases) by 0.5% (0.56%) in 2010, and by
about 1.3% (1.29%) in 2050.
The demographic transition is found to induce substantial changes in the sectoral
composition of output. Sectoral change—as measured by a change in the share of
sectoral output in total domestic output—ranges between −8.5% and +6.5% in year
2010 and between −23.5% and +20% in 2050. Expanding sectors are characterized
by higher growth rates of sectoral employment of labor and capital relative to the
economy-wide growth rates of labor and the capital stock. Accounting for structural
changes in life-cycle consumption that are due to age-specific preferences does not
affect the qualitative results of the benchmark model. At the aggregate level, the
quantitative effects are of minor significance. This suggests that the nature of the
economic transition is predominantly shaped by the negative labor supply shock.
At the sectoral level, however, demand-side induced effects stemming from age-
dependent consumer spending are found to be quantitatively important.
In order to evaluate the welfare consequences of the demographic transition the
following thought experiment is conducted: suppose households of different age and
type alive in 2003 live through the economic transition with changing factor prices
induced by the demographic change, how would their welfare change relative to a
situation without a demographic transition ? The answer is that young households
experience substantial welfare gains because they benefit from a future path of in-
creasing wage rates and do not suffer from too large losses of capital income on al-
ready accumulated financial wealth. In contrast, older asset-rich households tend to
gain less or even lose because of declining interest rates. From an intra-generational
perspective, those members of society for whom labor income constitutes a smaller
part of (future) resources, i.e. , households with relatively low labor productivity,
benefit less from the demographic transition.
In Germany, the demographic transition is projected to increase income inequal-
ity. The growing dispersion in household income is driven by a rise in capital income
inequality, and overcompensates the projected decrease in labor income inequality.
Labor income inequality decreases because a future path of increasing wages im-
plies a tendency towards a flatter labor supply profile over the life cycle. It is also
10 1 Introduction
2.1 Introduction
Over the past twenty years infinite horizon general equilibrium models with over-
lapping generations (OLG) have become an important tool for policy analysis, and
have been fruitfully applied in fields such as macroeconomics or public finance (see,
e.g., Auerbach and Kotlikoff (1987), and Kotlikoff (2000) for an overview). OLG
models naturally involve a large number of variables and equations that describe
the equilibrium behavior of economic agents. As a consequence, the development
of large-scale OLG models is often limited by the computational capacity of avail-
able numerical solution methods. In particular, models that exhibit a rich household
side including a variety of household-specific effects, a large number of heteroge-
neous households, and realistic agent lifetimes typically require “customized solu-
tion methods” which may be both costly to implement and difficult to validate.
This chapter1 develops a decomposition algorithm based on “off the shelf nu-
merical tools” for solving general equilibrium models with many households, of
which OLG models are a special case. The presented approach is primarily appro-
priate for computing equilibria for models in which the number of agents is so large
that simultaneous solution methods operating directly on the equilibrium system
of equations are infeasible due to the high dimensionality related to income and
household-specific effects. The “sequential recalibration” (SR) algorithm presented
here is based on the solution of a sequence of nonlinear complementarity problems2
although in special cases the same procedure may be implemented by solving a
1 This chapter has been jointly written with Thomas F. Rutherford. GAMS code for the presented
is convenient, robust, and efficient. A characteristic of many economic models is that they can
be cast as a complementary problem, i.e. given a function F : Rn −→ Rn , find z ∈ Rn such that
F(z) ≥ 0, z ≥ 0, and zT F(z) = 0 The complementarity format embodies weak inequalities and
complementary slackness, relevant features for models that contain bounds on specific variables,
e.g. activity levels which cannot a priori be assumed to operate at positive intensity. Such features
are not easily handled with alternative solution methods.
sequence of convex nonlinear programming problems. The main idea of the pre-
sented decomposition approach is to solve a market economy with many house-
holds through the computation of equilibria for a sequence of representative agent
economies. Typically, the sequence of prices and quantities converges to the true
equilibrium allocation.
The close connection between the allocation of a competitive market economy
and the optimal solution to a representative agent’s planning problem is well known
and widely cited in the economic literature. Negishi (1960) was the first to use an op-
timization problem to characterize equilibrium allocations in a general equilibrium
framework. Negishi’s original paper was primarily concerned with optimization as a
means of proving existence. Dixon (1975) developed the theory and computational
effectiveness of “joint maximization algorithms” for multi-country trade models.
Rutherford (1999b) presented the “sequential joint maximization algorithm” (SJM)
which provides a simple recursive version of Negishi’s method.
There are similarities between the SR and SJM algorithms. Both approaches
solve subproblems representing relaxations of the equilibrium conditions. The SJM
algorithm ignores consumer budget constraints but retains details of consumer de-
mand systems. The SR algorithm employs a yet looser representation of individual
consumer’s demand systems by omitting both income constraints and global prop-
erties of the individual utility functions. The omission of global characteristics of
preferences simplifies the model but can hinder convergence. The appropriateness
of the proposed solution method therefore depends on the characteristics of the un-
derlying model.
The decomposition approach can be useful for the computation of equilibria in
large-scale general equilibrium models with many households. There are many eco-
nomic questions for which heterogeneous agent models have to be used to provide
answers, and an increasing amount of research employs frameworks that allow for
intra-cohort heterogeneity in an OLG setup.3 We believe that the presented approach
can be beneficial for a wide range of economic applications, in particular within the
class of OLG models, for the following reasons. First, by significantly reducing the
computational overhead of the numerical problem at hand this method facilitates
the development of OLG models which feature a complex and rich household side.
This strengthens the microfoundation of the models in general and allows to an-
alyze in detail intra- and intergenerational distributive consequences of economic
policy. Second, the presented approach enables to solve OLG models that include a
“realistic” number of households within each age group since the number of house-
holds in the model more closely corresponds to the number of observational units
available from household survey data. This approach avoids relying on some ad-
3 For instance, Conesa and Krueger (1999), Kotlikoff et al. (1999), and Huggett and Ventura (1999)
investigate the intra-cohort distributive and welfare consequences of social security reform. Fehr
(2000) looks at pension reform during the demographic transition in the case of Germany. Ventura
(1999) explores the general equilibrium impact and associated distributional consequences of a rev-
enue neutral tax reform, and Jensen and Rutherford (2002) analyze the intra- and intergenerational
welfare effects of fiscal consolidation via debt reduction. This chapter concentrates on applica-
tions within the Auerbach-Kotlikoff OLG framework. For a general discussion of economies with
heterogeneous agents, see, e.g., Rios-Rull (1995).
2.1 Introduction 13
hoc aggregation of household groups, and thereby helps to enhance the empirical
basis of the model. Third, and more generally, the method can also be effectively
applied to OLG models which display a high dimensionality deriving from sources
other than the household side. Potential applications may include multi-sectoral and
multi-country models, or models which incorporate a detailed government sector.
In addition, the present contribution adds to the recent and growing body of re-
search that deals with the integration of macro and microsimulation models, the
“micro-macro” approach to modeling (Bourguignon and Spadaro, 2006). This strand
of literature aims to combine the strengths of both the computable general equilib-
rium (CGE) paradigm and microsimulation models. While CGE models have be-
come standard tools of quantitative policy assessment in the last twenty years, two
major criticisms are their reliance on the concept of the “representative agent” and
their usage of unclear aggregation procedures. The virtue of the microsimulation ap-
proach, on the other hand, is to replace representative agents with “real households”
as observed in standard household surveys. This, however, is typically achieved at
the cost of ignoring general equilibrium effects that are essential for policy analysis.
The simplest link between economy-wide modeling and the microsimulation ap-
proach proceeds top down, i.e. simulated policy changes obtained from an aggregate
representation of the economy are passed down to a microsimulation module, as,
e.g., in Bourguignon et al. (2005) and Bourguignon and Spadaro (2006). The princi-
pal weakness of the top-down approach is the absence of feedback effects from the
micro to the macro level. Relatively few studies have attempted to fully integrate
both approaches, most of them by means of employing an iterative strategy between
the microsimulation and the CGE model (Cockburn and Cororatona (2007), Savard
(2003, 2005), Arntz et al. (2006)). Also belonging to this class of models, Rutherford
and Tarr (2008) applied the SR algorithm to a large-scale, static general equilibrium
model with 25 sectors and 53,000 households to assess the poverty effects of Rus-
sia’s WTO accession. All of these studies, however, are concerned with applications
in a static framework. Clearly, covering complex behavioral responses and potential
general equilibrium and macroeconomic effects in a dynamic setup is essential for
many policy issues.4
The present chapter develops a computational technique that allows to fully inte-
grate a comprehensive system of OLG households, that exhibits a substantial degree
of intra-cohort heterogeneity, into a generic Auerbach-Kotlikoff model. We show
that the positive experience of the SR algorithm for large-scale static models car-
ries over to dynamic applications and demonstrate its effectiveness for solving OLG
models with a large number of heterogeneous households. To find the equilibrium
transition path of the OLG economy, the presented algorithm solves a sequence of
“related” Ramsey optimal growth problems where the system of overlapping gener-
ations is replaced by an infinitely-lived representative agent. An iterative procedure
4 Available models tend to concentrate on some specific behavior, abstracting from other impor-
tant components of the demo-economic life cycle. For instance, Townsend (2002), Townsend and
Ueda (2003) concentrate on saving/investment behavior under uncertainty and in different financial
market environments.
14 2 Computation of Equilibria in OLG Models with Many Heterogeneous Households
between the macro and micro model is employed based on the successive recalibra-
tion of preferences of this artificial representative agent.
In order to characterize limitations of the algorithm, local convergence theory
for a simple exchange economy due to Scarf (1960) is developed and a number
numerical analyses are carried out to examine global conditions under which the
SR algorithm may fail to converge. We show that conditions for local stability of
the adjustment process reduce to those of a Walrasian price tâtonnement process,
thus SR belongs to a large class of algorithms commonly used in computational
economics which are robust and efficient, yet fail to provide global convergence.
The presented counterexample illustrates that the SR algorithm may be ill-suited for
applications in which there are significant income effects.
After having characterized limitations of the technique, this chapter explores the
algorithm’s performance when applied to large-scale OLG models. To this end, a
prototype Auerbach-Kotlikoff model is considered which includes up to 2000 het-
erogeneous households within each generation differing with respect to labor pro-
ductivity over the life cycle and other behavioral parameters. The performance of
the SR algorithm is compared to a simultaneous solution method as suggested by
Rasmussen and Rutherford (2004). We find that the proposed algorithm can pro-
vide improvements in both efficiency and robustness. It is furthermore demonstrated
that the decomposition algorithm can routinely solve high-dimensional OLG mod-
els which are infeasible for conventional solution methods.
Lastly, it is important to emphasize that the decomposition method is inadequate
for approximating equilibria in OLG economies that are generically Pareto-inferior,
i.e. models in which the economy’s growth rate exceeds the real interest rate (see,
e.g., Diamond (1965) and Phelps (1961)). For the given model, this corresponds to
a situation where population growth dominates discounting. In such circumstances
there is no social planner’s problem which corresponds to the OLG demand system.
Whether this significantly limits the relevance and scope of the presented approach
ultimately is an empirical question. Empirical evidence suggests that the incapacity
of the method to deal with dynamically inefficient equilibria is of minor practical
significance.5
The rest of the chapter is organized as follows. Section 2.3 introduces the SR
algorithm for the case of a static economy and illustrates its basic logic by means
of graphical analysis. Section 2.4 investigates a model where convergence of the SR
algorithm fails if income effects are relatively strong. Section 2.5 demonstrates that
the algorithm can be effectively applied to solve Auerbach-Kotlikoff OLG models.
Furthermore, the performance of the algorithm is compared to computational expe-
rience from an integrated simultaneous solution method. Section 2.6 concludes.
5 Abel et al. (1989) find that for the US economy the condition for dynamic efficiency seems to be
satisfied in practice. Similarly, under the weak assumption that rates of return are ergodic, Barbie,
Hagedorn and Kaul (2004) reach the conclusion that the US economy does not overaccumulate
capital. By means of numerical analysis, Larch (1993) suggests that in the Auerbach-Kotlikoff
framework rather implausible values of the pure rate of time preference, the intertemporal elasticity
of substitution or the population growth rate are required to obtain non Pareto-optimal market
solutions.
2.2 Calibration in CGE models 15
In order to facilitate the subsequent description of the algorithm and to clarify the al-
gebraic forms used in the associated computer programs, this section reviews some
fundamental aspects of calibration which underly most computable general equilib-
rium (CGE) models.
Calibration refers to the process of selecting values of model parameters which en-
sure that the model’s reference equilibrium is consistent with given data. Such data
are typically obtained in the form of a social accounting matrix for a given base
year. CGE models are based on parametric forms which describe technology and
preferences. The most common functional form used in empirical applications is
the constant-elasticity-of-substitution (CES) function. A CES utility function can
be written as: ! 1/ρ
n
ρ
U(C) = ∑ αi Ci (2.1)
i=1
where pi is the price of consumption good i, and M is consumer income. The con-
sumer problem in closed-form can be solved, obtaining demand functions:
αiσ M p−σi
Ci (p, M) = (2.3)
∑i0 αiσ0 p1−σ
i0
where p denotes the price vector, and i 6= i0 . The calibration of preferences involves
inverting this demand function to express the function parameters in terms of an
observed set of prices and demands. If a consumer chooses to consume quantities
Ci when commodity prices are pi , it may be concluded that the share parameters
have to be given by:
1−ρ
α i = λ pi C i (2.4)
16 2 Computation of Equilibria in OLG Models with Many Heterogeneous Households
where λ > 0 is an arbitrary scale factor6 , and the elasticity parameter, ρ, is exoge-
nously specified. It is helpful to think of the share and scale parameters as cali-
brated values, determined by an agent’s observed choices in a reference equilibrium,
whereas the elasticity parameters are “free parameters” which are typically drawn
from econometric estimates of the responsiveness of demand or supply to changes
in relative prices. In traditional applied general equilibrium models, the reference
quantities Ci and prices pi are based on a benchmark equilibrium data set.
In applied work it may be convenient to work with a different yet equivalent form of
the CES utility function (Rutherford, 1995a). The calibrated share form is based on
the observed quantities, prices and budget shares. In computational applications the
calibrated form is preferable because it provides a simple parameter and functional
check that is independent from second-order curvature. Normalizing the benchmark
utility index to unity, the utility function can be written as:
" ρ #1/ρ
Ci
U(C) = ∑ θi (2.5)
i Ci
in which:
pi Ci
θi = (2.6)
∑i0 pi0 Ci0
is defined as the benchmark value share of good i. Similarly, one can express the
unit expenditure function as:
" 1−σ #1/1−σ
pi
e(p) = ∑ θi , (2.7)
i pi
where M = ∑i pi Ci .
6 The consumer maximization problem is invariant with respect to positive scaling of U, hence the
One can think of the demand function given here as a second-order Taylor approx-
imation to the “true” demand function based on an observation of the true function
at the reference point. At that point, Ci corresponds to a “zeroth order approxima-
tion” to the utility function, pi corresponds to the “first order approximation”, and
the “free parameter” σ controls the second (and higher) order properties of prefer-
ences. The benchmark prices correspond to the marginal rate of substitution—the
slope of the indifference curve—at C. As long as one remains in the neighborhood
of p, the elasticity parameter σ only plays a minor role, and calibrated demand is
determined largely by C and p.
This section presents the decomposition method by which a market economy with
many heterogeneous households may be solved through the computation for equi-
libria for a sequence of representative agent economies. While the primary interest
is in dynamic models, it is advantageous to introduce the algorithm for the case of
a two-sector static economy in which one can provide a graphical description that
serves to illustrate its basic logic.
2
s.t. ∑ pi chi = Mh . (2.10)
i=1
where the utility function is written in calibrated share form. θih and chi denote the
benchmark value share and the benchmark consumption of good i for household h.
Households are heterogeneous with respect to θih , ρ h , K h , and Lh .
Furthermore, there is a single representative firm which uses capital and labor
services to produce two consumption goods Xi , i, j = 1, 2, according to a constant
returns to scale production function Xi = fi (K, L), where K = ∑h K h and L = ∑h Lh .
All goods and factor markets are perfectly competitive.
18 2 Computation of Equilibria in OLG Models with Many Heterogeneous Households
The main challenge for computing equilibria in a setup where H is very large is
dimensionality. Typically, conventional simultaneous solution methods that operate
directly on the equilibrium system of equations are infeasible. The proposed algo-
rithm decomposes the corresponding numerical problem into two parts and thereby
effectively manages to reduce its dimensionality. The general equilibrium of the un-
derlying economic model is approximated by computing equilibria for a sequence
of representative agent economies. In each iteration, first a general equilibrium rep-
resentation of the underlying economic model is solved in which the household side
is replaced by a single representative agent (RA). The second subproblem then con-
sists of solving a partial equilibrium relaxation of the original model that retains
the full structure of the household demand system. Given equilibrium prices from
the previous solution of the RA economy, it is possible to compute optimal choices
for each of the “real” households. In a next step, and to create a basis for succes-
sive iterations, the preferences of the “artificial” RA are recalibrated such that given
candidate equilibrium prices the RA choices replicate aggregate household choices.
The key departure from the routine use of calibration in the decomposition al-
gorithm is the idea that the calibration of preferences occurs more than once. The
first iteration of the algorithm is based on observable benchmark data, but in sub-
sequent iterations the preferences of the RA are sequentially recalibrated to values
determined in the iterative process. For the case of the static economy as described
above, the SR algorithm involves the following steps:
where k is an iteration index. Factor endowments of the RA equal the sum of re-
spective factor endowments across all households. To initialize the RA economy at
a consistent data point, the data set has be constructed such that the RA model and
the household model share the same optimal consumption quantities in the initial
benchmark. This is achieved by setting:
2.3 CGE with Many Households: A Decomposition Approach 19
X2
A B
P0 P1
X1
Fig. 2.1 Solution to the initial representative agent model (Step 0 and Step 1)
H
0
Ci = ∑ chi (2.12)
h=1
0
pi C i
Θi0 = 0
, i 6= i0 (2.13)
∑i0 pi0 Ci0
0
where Ci and Θi0 denote initial consumption by the RA and the aggregate value
share for good i in iteration k = 0, respectively.
This initial consumption point of the RA in the benchmark equilibrium is rep-
resented by point A in Figure 2.1 where initial goods prices are denoted by P0 .
Benchmark prices and an arbitrary elasticity are used to extrapolate preferences in
the neighborhood of the benchmark point to the global preferences of the RA, as
indicated by the indifference curve which is tangent to the benchmark budget con-
straint at point A. The key limitation of the RA model on the demand side is that
the “community indifference curve” represented by this indifference curve does not
truthfully portray the response of household demand to a comprehensive change in
both goods and factor prices.
The solution to the RA model in the first iteration of the algorithm is illustrated in
Figure 2.1. As depicted here, the assumed exogenous policy shock has led to an in-
crease in factor earnings, and a reduction in the relative price of X1 as compared with
20 2 Computation of Equilibria in OLG Models with Many Heterogeneous Households
X2
A B
P0 P1
X1
X2 . This new price situation is denoted by P1 . The RA model, based on the assumed
community indifference curve and the associated change in factor and commodity
prices returns point B as the optimal consumption point.
In the solution program equilibrium prices are read from the RA model and evaluate
the household demand vector. This produces a different point on the same budget
constraint (see Figure 2.2). The household demand model is based on compensated
demand functions so the aggregate budget constraint for the household demand sys-
tem is equivalent to the budget constraint which applies to the RA. Point C corre-
sponds to the aggregate demand which results from solving the individual household
optimization problems. The extent to which C differs from B depends on both the
difference in implicit substitution elasticities and differences in income effects.
The next step in the algorithm consists of specifying a new set of preferences for
the RA model. The algorithm is termed “sequential recalibration” on the basis of
this idea. After having solved one RA model, a new RA model is constructed based
on a set of preferences which are locally calibrated to the aggregate consumption
quantities at point C and the associated relative prices. This ensures that given prices
P1 the optimal consumption point of the new RA, point C in Figure 2.3, is consistent
with the aggregated choices by households. Preferences of the RA in iteration k,
2.3 CGE with Many Households: A Decomposition Approach 21
X2
A B
P0 P1
X1
U k (C) in (2.11), are based on household demands at the prices returned in iteration
k − 1:
k
Ci = ∑ chi (pk−1 , Mhk−1 ) (2.14)
h
in which chi (pk−1 ) is the demand for good i by household h evaluated at the candidate
price vector from the previous iteration k − 1, and where factor income of household
h in iteration k, Mhk (pk ), is a function of prices in iteration k. Likewise, value shares
in U k (C) are updated to:
pk−1
i ∑h chi (pk−1 , Mhk−1 )
Θik = (2.15)
∑ j pk−1
j ∑h chj (pk−1 , Mhk−1 )
where θik is the aggregate value share of good i at iteration k − 1. The indifference
curves tangent at A and B are based on identical preferences, but the indifference
curve tangent at point C is based on a new set of community indifference curves,
hence it may intersect the indifference curves based on RA utility in the previous
iteration of the algorithm. Note that the preferences of the “real” households remain
unchanged throughout the entire iteration process.
22 2 Computation of Equilibria in OLG Models with Many Heterogeneous Households
X2
C
D
A B
PPF PPF' P1 P2
X1
Fig. 2.4 Iterative adjustment
Iterative Adjustment
When the RA model is recalibrated at point C, both the representative agent and
all households are in equilibrium at C with prices P1 , but at these prices firms will
only supply quantities given by point B. Hence, due to inconsistency with the sup-
ply side of the model there is a general disequilibrium. To illustrate this idea, it is
convenient to portray the supply side of the economy by a production possibility
frontier (PPF). Assume that the policy shock produces an expansion in the PPF (to
PPF 0 ) and a substantial change in relative prices from point A to B. The next step
in the solution program is to resolve for a general equilibrium of the new RA model
with recalibrated preferences at point C. Point C in Figure 2.4 becomes therefore
interpreted as point A in the next iteration. The solution of this RA model is then
characterized by a new optimal consumption point, here depicted by point D, and
prices P2 .
Subsequent iterations involve carrying out Steps 1 to Steps 3 (Step 0 initializes
the solution procedure). The process is stopped if some convergence metric, e.g., the
1-norm of the difference between the price vectors from one iteration to the next, is
satisfied. Note that subsequent iterations of the algorithm only involve refinements
of the demand system and result in much smaller changes in relative prices, as indi-
cated here by the change from C to D as compared with A to B.
2.4 Convergence Theory 23
This section evaluates the performance of the algorithm for an economy in which
the exact equilibrium is known and where the computed allocation can be compared
to the true equilibrium allocation. Local convergence theory for the proposed algo-
rithm is developed and conditions under which the adjustment process may fail to
converge are identified.
The example is due to Scarf (1960) who considers a pure exchange economy
with an equal number of n consumers and goods. Consumer h is endowed with one
unit of good h and demands only goods h and h + 1. Let di,h denote demand for
good i by consumer h. Preferences are represented by CES utility functions with the
following structure:
σ σ σσ−1
σ −1 σ −1
Uh (d) = θ dh,h + (1 − θ ) dh+1,h . (2.16)
Scarf (1960) demonstrates that this economy has a unique equilibrium in which all
prices are equal to unity.7
7 See Lemma 1 and Lemma 2 (Scarf, 1960, p.164). The parameters of this utility function corre-
1 b
spond to Scarf’s parameters a and b (Scarf, 1960, p.168) as: σ = 1+a and θ = 1+b .
24 2 Computation of Equilibria in OLG Models with Many Heterogeneous Households
∂ ζi ( p(C))
∂ ξi ∂ Ci
i= j
ξi j ≡ = . (2.19)
∂ C j ∂ ζi ( p(C)) i 6= j
∂Cj
∗
If all principal minors of ∇ξ (C ) = [ξi j ] are negative, the adjustment process is
∂ ζ ( p(C))
locally convergent. If, however, i∂ C > 0, the process is locally unstable. When
i
an equilibrium is unique and the process is uninterrupted, then local stability implies
global instability.
For this model, the tâtonnement price adjustment process is unstable (in the case
θ 1
n = 3) when 1−θ > 1−2σ (Scarf, 1960). In the following, it is shown that the same
condition implies instability for the C-adjustment process of the SR algorithm. Fur-
thermore, it is shown (numerically) that while the tâtonnement and SR price adjust-
ment processes are locally identical, they may be quite different at points in the price
space that are sufficiently far away from the equilibrium.
Given the special structure of ζi p(C) , one can write:
θ pi (1 − θ ) pi
di,i = 1−σ σ
, di+1,i = . (2.22)
ei pi e1−σ
i pσi+1
∂ ζi ( p(C))
Evaluating ∂ Ci
at p∗ = 1, yields
∂ ζi p(C) ∂ pi
(2σ − 2)θ 2 + (3 − 2σ )θ − 1
=
∂ Ci ∂ Ci
∂ pi+1
+ (−θ (1 − θ )(1 − σ ))
∂ Ci
∂ pi−1
+ (−θ (1 − θ )(1 − σ ) + 1 − θ ) . (2.23)
∂ Ci
Ci
ζei p,C = −1 (2.25)
∑i0 αi0 p1−
i0
σe
pσie
Ci
with αi0 = and where σe denotes the elasticity of substitution for the repre-
∑ i0 C i0
∗
sentative agent. In order to evaluate ∇ p ζe at C , we make a first-order Taylor series
expansion:
∇ p ζe p,C d p + ∇C ζe p,C d C = 0 (2.26)
which gives:
dp
= −∇−1
p ζ ∇C ζ .
e e (2.27)
dC
∗
Evaluating gradients at p∗ = C = 1 yields:
−(2σe +1) −(1−σe ) −(1−σ e)
3 3 3
−(1−σe ) −(2σ
e +1) −(1−σ e)
∇ p ζe = (2.28)
3 3 3
−(1−σe ) −(1−σ e ) −(2σ
e +1)
3 3 3
e −1 σ
e +2 σ
σ e −1
3σ
e 3σ
e 3σ
e
−∇−1
p ζ =
e e −1 σ
σ e −1
e +2 σ
(2.29)
3σ
e 3σ
e 3σ
e
e −1 σ
σ e −1 σ
e +2
3σ
e 3σ
e 3σ
e
100
∇C ζe = 0 1 0 . (2.30)
001
Hence, in the neighborhood of the equilibrium:
∂ pi (C) σe + 2 ∂ pi (C)
= > 0, = 0. (2.31)
∂ Ci 3σe ∂ Cj
From (2.23) it therefore follows that the adjustment process in C is locally unstable
if:
(2σ − 2)θ 2 + (3 − 2σ )θ − 1 > 0 (2.32)
which is equivalent to the condition for instability of a simple price tâtonnement
adjustment process as demonstrated by Scarf (1960, p.169).
26 2 Computation of Equilibria in OLG Models with Many Heterogeneous Households
p3
p1 p2
Although the local behavior of the price tâtonnement and the SR algorithm adjust-
ment processes are identical, they produce different search directions away from
a neighborhood of the equilibrium. This is apparent in Figure 2.5 where the two
vector fields are superimposed. Only local to the equilibrium where price effects
dominate income effects, do the two fields coincide exactly, as indicated by (2.32).
As one moves further away from the center of the simplex, the vector fields become
more divergent. It is found that there are cases in which the SR algorithm does not
converge even though the price tâtonnement is globally stable. This convergence
failure is a manifestation of the simplifying nature of the adjustment process. By
solving a sequence of representative agent economies the SR algorithm omits both
income constraints and global properties of the individual utility functions. While
the omission of global characteristics of preferences reduces the dimensionality of
the model significantly, this may at the same time hinder convergence.
2.4 Convergence Theory 27
16
2.00 12
8
1.75
4
1.50 non-convergent
1.25
s 1.00
0.75
0.50
0.25
8 For both parameters, a grid resolution of 0.05 is chosen, σe is set to one, and a maximum of 1000
iterations is allowed for. The adjustment process is said to converge if the 1-norm of differences
between a computed price vector and the equilibrium point drops below some metric δ , i.e. kpi −
p∗i k1 < δ , where p∗i denotes the analytical equilibrium solution. We set δ = 0.01.
28 2 Computation of Equilibria in OLG Models with Many Heterogeneous Households
2.5.1.1 Households
9 The example is an adapted version of the production model presented in Rasmussen and Ruther-
ford (2004). A closed economy version of their model with intra-cohort heterogeneity is considered.
While a single-sector model is investigated here, the logic can be readily extended to a multisectoral
framework.
10 ω is a constant income scaling factor which is determined in the initial calibration procedure to
reconcile household behavior with the aggregate benchmark data. For more details see Rasmussen
and Rutherford (2004).
2.5 OLG Models with Many Households 29
2.5.1.2 Firms
There is a single representative firm which in each period t uses capital and labor
services to produce a single output good Yt according to a linearly homogeneous
11 We therefore implicitly assume that the government has no outstanding debt at period zero. A
situation with non-zero initial government debt slightly complicates the calibration procedure but
is conceptually straightforward (see Rasmussen and Rutherford (2004)).
30 2 Computation of Equilibria in OLG Models with Many Heterogeneous Households
production function Yt = F(Kt , Lt ). All goods and factor markets are perfectly com-
petitive.
2.5.1.3 Government
The government agent collects revenue from levying taxes on consumption, and
on capital and labor income. Tax revenue is spent on government expenditure (Gt )
and on total transfers to households (Tt = ∑tg=t−N ∑H h=1 ζg,h,t ). We assume that the
consumption tax rate (τtc ) adjusts such that the government budget is balanced on a
period-by-period basis:
where τtr and τrl are the net tax rates on capital and labor income, respectively.
t H
It = ∑ ∑ ig,h,t (2.36)
g=t−N h=1
t H
Kt = ∑ ∑ kg,h,t (2.37)
g=t−N h=1
t H
Ct = ∑ ∑ cg,h,t . (2.38)
g=t−N h=1
The law of motion for the aggregate capital stock is given by:
Finally, the single output good may be used for household consumption, investment,
or government consumption implying the following condition for balance between
aggregate supply and demand:
F(Kt , Lt ) = Ct + It + Gt . (2.40)
2.5 OLG Models with Many Households 31
Step 1: Step 2:
Solve related Ramsey Evaluate household
growth problem demand functions
Step 3:
Recalibrate preferences
of the Ramsey agent Partial equilibrium
quantity choices
As the “related” Ramsey optimal growth problem, we define a model of the under-
lying OLG economy in which the system of overlapping generations is replaced by
32 2 Computation of Equilibria in OLG Models with Many Heterogeneous Households
Kt+1 ≤ (1 − δ ) Kt + It
Lt ≤ Ωtk
Ct , Lt ≥0
K0 ≤Ψ
KT +1 = K̂T +1 (2.41)
where Ct , Lt , Zt , Kt , It , and Ωtk now denote consumption, leisure time, full consump-
tion, the capital stock, investment, and the time endowment by the Ramsey agent,
respectively, and where σb is the intertemporal elasticity of substitution. k denotes
an iteration index.12
The initial capital stock in the Ramsey economy is given by the aggregate capital
stock of the OLG economy in year zero:
0 H
Ψ= ∑ ∑ kg,h,0 . (2.42)
g=−N h=1
12 Note that the nested lifetime utility function U(Zt ) is written in calibrated share form. We mono-
tonically transform preferences in (2.33) to obtain a linear homogenous CES representation. This
does not alter the underlying preference orderings and hence optimization yields the same demand
functions.
2.5 OLG Models with Many Households 33
2.5.2.2 Initialization
Ramsey
0 T
OLG
0 T– N+1 T T+N
terminal post-terminal
generations periods
The second step of the algorithm solves a partial equilibrium relaxation of the un-
derlying OLG economy which retains full details of the household demand system
but ignores general equilibrium effects. Hence, any interactions via commodity and
factor markets and with the production side of the economy are suppressed. Given
equilibrium prices from the previous solution of the “related” Ramsey growth prob-
lem, we evaluate demand functions for each generation and type that originate from
the set of household problems in (2.33).
In order to obtain a good approximation of the underlying OLG economy, it
is necessary to compute optimal household demand for all households and types
in each period of the numerical model that runs from t = 0, . . . , T . This infor-
mation then forms the basis for the recalibration of preferences of the Ramsey
agent in the subsequent step of the algorithm. A complication arises for periods
T − N + 1 ≤ t ≤ T in which generations are born that live beyond T (henceforth
called terminal generations). Figure2.8 illustrates this issue. To compute the optimal
decision profiles of these agents, it is essential to account for their behavior over the
full life cycle. With the last cohort of households being born in period T , this means
that there are N post-terminal periods that have to be included in the analysis, which
we denote by tˆ = T, . . . , T + N. We resolve this issue by employing a steady-state
closure rule which postulates that the economy has reached a steady state by pe-
riod T − N + 1. This additional restriction is not binding if T is chosen sufficiently
large.13 Exploiting this fact, prices for post-terminal periods can be inferred from
the following steady-state projection:
13 The specific choice of T depends on the nature of the policy shock that is considered. In the
xTk
xtkˆ = (2.50)
(1 + r∞ )tˆ−T
where eg,h,t (·) denotes the unit expenditure function for zg,h,t . For future reference,
let pg,h,t denote respective benchmark prices. In (2.50), r∞ = pcT −1 /pcT − 1 defines
the endogenous (steady-state) interest rate in the terminal period.
Finally, the lifetime income of generation g and type h, evaluated at candidate
prices from iteration k, is given by:
g+N
k
Mg,h = ∑ πg,h,t pkl,t ωg,h,t + pky,t ζg,h,t + pkr,0 kg,h,g . (2.52)
t=g
A similar formula applies to the lifetime income of terminal generations where for
post-terminal periods projected prices according to (2.50) are used. For future refer-
ence, let M g,h denote the lifetime income at benchmark prices.
We are now in a position to compute household demand. In order to reduce com-
putational complexity, we solve the dual problem making use of formulas (2.7)-(2.9)
(see the Appendix). Let eg,h (pkg,h ) denote the expenditure function for a unit of ug,h
which—given the specific structure of preferences14 —can be constructed using the
vector of prices for the full consumption good, pkg,h (including projected prices).
Indirect utility can then be written as:
k
Mg,h
Vg,h (pkg,h , Mg,h
k
)= ∀g, ∀h. (2.53)
eg,h (pkg,h ) M g,h
Applying Roy’s identity, optimal household demand (in the context of the partial
equilibrium relaxation) for full consumption, goods consumption and leisure, re-
spectively, evaluated at the candidate price vector pkg,h , are updated in each iteration
according to:
14 Note that in the given case of homothetic preferences, the unit expenditure function conveys all
information concerning the underlying preferences.
36 2 Computation of Equilibria in OLG Models with Many Heterogeneous Households
!σh
eg,h (pkg,h ) pg,h,t
zg,h,t (pkg,h , Mg,h
k
) = zg,h,t Vg,h (pkg,h , Mg,h
k
) (2.54)
pkg,h,t
!σν
pkg,h,t pt
cg,h,t (pkg,h , Mg,h
k
) = cg,h,t zg,h,t (pkg,h , Mg,h
k
) (2.55)
pkc,t pg,h,t
!σν
pkg,h,t pt
`g,h,t (pkg,h , Mg,h
k
) = `g,h,t zg,h,t (pkg,h , Mg,h
k
) . (2.56)
pkl,t πg,h,t pg,h,t
The last step in each iteration is to construct a new Ramsey optimal growth problem
by recalibrating the Ramsey agent’s preferences based on optimal household choices
from the previous step. This is accomplished by updating level parameters in (2.41)
according to:
t H
k+1
Ct = ∑ ∑ cg,h,t (pkg,h , Mg,h
k
) (2.57)
g=t−N h=1
t H
k+1
Lt = ∑ ∑ `g,h,t (pkg,h , Mg,h
k
) (2.58)
g=t−N h=1
k k k
Z t = Ct + L t (2.59)
Thus, the newly constructed Ramsey optimal growth problem in iteration k + 1 con-
sists of solving (2.41) (subject to (2.42) and (2.43) with updated preference parame-
2.5 OLG Models with Many Households 37
ters and time endowment as defined by (2.57)–(2.62). This completes the description
of the algorithm.
The OLG economy presented above has no analytical solution. In order to evaluate
the algorithm, we therefore compare its performance to those of conventional simul-
taneous/direct solution methods. As a benchmark, we take a complementarity-based
approach as suggested by Rasmussen and Rutherford (2004).
The base case parametrization of the economy is as follows. Households live for
51 years or N = 50. We set r̄ = 0.05, γ = 0.01, δ = 0.07, νh = 0.8, β = 0.32, and α =
0.8. In the numerical analysis, we test the performance of the algorithm for a differ-
ent number of household types H and also allow for various degrees of intra-cohort
heterogeneity. For simplicity, we assume that σh , h = 1, . . . , H, are generated by ran-
dom draws from a uniform distribution defined over [σ , σ ]. Likewise, differences in
labor productivity are generated by randomly drawing ag,h from a uniform distribu-
tion with support a ≤ ag,h ≤ a, where the parameter
h ag,h enters the labor productiv- i
ity profile over the life cycle as: πg,h,t = exp 4.47 + ag,h (t − g) − 0.00067 (t − g)2 .
Furthermore, it is assumed that each type has equal size in the total population. The
values for the aggregate data including tax payments in the initial benchmark are
based on Input-Output tables for the U.S. economy in 1996 and are presented at the
top of the corresponding computer programs. We solve the model for T = 150 years.
Iteration 1 (= Ramsey)
Deviation from steady state (in %)
7.5
Iteration 2
Iteration 3
4
5
7
6
7
...
6.5
OLG = Iteration 34
6
0 20 40 60 80 100 120 140
Time period
for certain parameter values. In such cases, indeterminacy would manifest itself as sensitivity to the
truncation date. None of the models presented here are sensitive to T, provided that it is sufficiently
large. This and the general robustness of the models provide evidence that the equilibria are unique.
Kotlikoff (2000) reaches the same conclusion regarding the uniqueness of equilibria in the OLG
models he has been working with.
2.5 OLG Models with Many Households 39
Iteration 1
1 Iteration 2
OLG = Iteration 34
0.5
-0.5
-40 -20 0 20 40 60 80 100 120 140
Generation
In order to explore the capacity of the algorithm to solve large-scale OLG models,
we examine its performance for a different number of households and various de-
grees of intra-cohort heterogeneity. We look again at the effects of the tax reform
scenario as described above. Given the simple specification for the source of intra-
cohort differences, we vary the extent of heterogeneity, denoted by Γ , by changing
the support for the distributions from which σh and ag,h are drawn.16
Table 2.1 reports results from a series of runs where the number of households
within each generation is increased while holding fixed the degree of intra-cohort
heterogeneity. The quality of approximation is excellent (τk is around 10−4 ). As
the number of household types increases, the proposed decomposition procedure be-
come advantageous.17 Most importantly, it is shown that the algorithm can provide
improvements in robustness as compared to the benchmark simultaneous solution
method which quickly becomes infeasible for models in which H ≥ 100.
To examine the performance of the algorithm in the presence of a substantial
degree of heterogeneity among households, we report results for different configura-
16 We consider the following sets of choices for {(σ , σ ), (a, a)} ordered by their implied
degree of heterogeneity: Γ1 = {(1.00, 1.50), (0.2, 0.3)}, Γ2 = {(1.00, 1.50), (0.2, 0.4)}, Γ3 =
{(0.25, 0.75), (0.2, 0.3)}, Γ4 = {(0.25, 0.75), (0.2, 0.4)}, Γ5 = {(0.25, 1.25), (0.2, 0.3)}, Γ6 =
{(0.25, 1.25), (0.2, 0.4)}, Γ7 = {(0.25, 2.00), (0.2, 0.3)}, Γ8 = {(0.25, 2.00), (0.2, 0.4)}.
17 All reported running times refer to an implementation on a Dual Core 2 GHz processor machine.
40 2 Computation of Equilibria in OLG Models with Many Heterogeneous Households
0.015
0.012
1-norm of price changes
0.009
0.006
0.003
0
1 5 10 15 20 25 30
Iteration
tions of Γ . We set H = 50 so that the benchmark solution method is feasible and the
calculation of approximation errors is available. Not surprisingly, the approximation
quality of the method is decreasing with the degree of heterogeneity. Overall, the
quality of approximation is still very good: computed prices fall within a reasonably
small interval around the true equilibrium price path (τ k is around 10−5 − 10−3 ).
Motivated by the discussion of the potential convergence failure of the SR al-
gorithm in the presence of significant income effects (Section 2.4), we conduct a
number of sensitivity analyses for behavioral parameters governing intra-period and
intertemporal substitution and income effects (see Table 2.3). Looking first at the
intra-period dimension, we find that combinations of too small ν and α can pose
serious problems for the decomposition approach. Although the search process is
terminated within a modest number of iterations, both approximation measures in-
dicate a rather poor quality of approximation for α ≤ 0.5. If α is too small, the
equilibrium behavior over the life cycle of a household displays periods with zero
labor supply in old ages, i.e. there is endogenous retirement. This happens because
the shadow price of time exceeds the market wage rate. In the presence of such
corner solutions, it is harder to portray the choices of OLG households by using a
representative agent which in turn explains why the approximation error increases.
As for the role of intertemporal income effects, we do not experience problems
of convergence or a poor quality of approximation (results not shown). However,
the speed of convergence (in terms of the number of iterations required for con-
vergence) is the slower, the larger is σh . This finding indicates that income effects
2.6 Concluding Remarks 41
Note: Figures in parentheses denote running time of the decomposition algorithm expressed as a
fraction of the running time as required by the benchmark simultaneous solution method. A “×”
indicates infeasibility of the simultaneous solution method.
Note: Figures in parentheses denote running time of the decomposition algorithm expressed as a
fraction of the running time as required by the benchmark simultaneous solution method.
This chapter develops a decomposition approach which can be applied to solve high-
dimensional static and dynamic general equilibrium models with many households.
We demonstrate its effectiveness for computing equilibria in large-scale OLG mod-
els which are infeasible for conventional simultaneous/direct methods. We find that
the proposed algorithm provides an efficient and robust way to approximate gen-
eral equilibrium in models with a large number of heterogeneous agents if income
effects remain sufficiently weak. The appropriateness of the solution method there-
42 2 Computation of Equilibria in OLG Models with Many Heterogeneous Households
Note: Figures in parentheses denote running time of the decomposition algorithm expressed as a
fraction of the running time as required by the benchmark simultaneous solution method.
fore depends on the characteristics of the underlying model and the nature of the
implemented policy shock.
We believe that the approach can be beneficial for a wide range of economic
applications. In particular, it is advantageous for modeling tasks which necessitate
to economize on the dimensionality of the corresponding numerical problem. Po-
tential applications may include multi-country and multi-sectoral OLG models, and
analyses of relevant policy issues—such as, e.g., population aging, trade policy, and
poverty—which require detailed account of the distributional effects on a household
level while at the same time taking into account general equilibrium effects. More-
over, the decomposition approach may prove useful for the further development of
fully-integrated static and dynamic microsimulation models that incorporate the es-
sential macroeconomic linkages required for a comprehensive policy analysis.
Chapter 3
Trade Liberalization and Global Demographic
Change: A Quantitative Assessment
3.1 Introduction
1 There is a vast literature which tries to assess whether trade liberalization is beneficial. Among
those which tend to find a positive effect, are, e.g., Edwards (1993), Frankel and Romer (1999),
7 5
C h in a
E u ro p e
W o r k in g - a g e p o p u la tio n r a tio ( in % ) F +G + I
7 0
J a
p a n
N o r th A m e r ic a
6 5
6 0
5 5
5 0
0
2 0 1 0 2 0 2 0 2 0 3 0 2 0 4 0 2 0 5 0
Y e a r
Fig. 3.1 Unsynchronized global aging patterns
Note: The working-age population ratio is defined as the population aged between 15-64 as a frac-
tion of the total population. ’F+G+I’ denotes the (unweighted) country average for France, Ger-
many, and Italy. Source: Own calculations based on United Nations, World Population Prospects:
The 2004 Revision. Medium projection variant.
The second innovation of this chapter is to analyze the distribution of gains from
trade, both across time and generations of households and within each age group.
We are not aware of any similar attempt in the literature. Besides the heterogeneity
of age among overlapping generations, the model also features a substantial amount
of intra-cohort heterogeneity: household types differ with respect to age-specific la-
bor productivity and other key parameters governing the relative magnitudes of intra-
and intertemporal income and substitution effects. Third, and in terms of computa-
tional methodology, we employ the decomposition algorithm proposed in Chapter 2
that approximates the equilibrium allocation of the OLG economy by solving a se-
quence of appropriately chosen two-country Ramsey growth models. Despite of the
large dimensionality of the numerical problem at hand—resulting from the system
of overlapping generations and the two-country setup—this decomposition method
makes it possible to compute the transitional and steady state equilibrium effects of
all endogenous model variables in response to a demographic shock.
The results of this chapter are as follows. Regionally unsynchronized demograph-
ics patterns lead to the emergence of international trade flows. This result is based on
a simple yet convincing model mechanism. In a nutshell, an economy with a grow-
ing fraction of older people is bound to increase its accumulation of capital and
bound to decrease its aggregate labor supply. Capital becomes abundant relative to
labor because life-cycle savings and labor supply behavior of overlapping genera-
tions entails that older households tend to hold more assets and work less compared
to younger ones. Factor prices change accordingly to reflect the varying scarcity of
factors of production, i.e. wages go down and real interest rates increase. This in turn
creates differences in production costs of commodities that are produced with dif-
ferent factor intensities. The induced sectoral change is characterized by a relative
contraction (expansion) of the labor-intensive (capital-intensive) sector. Under au-
tarky, differences in demographic patterns across regions translate into international
differences in the relative abundance of factors of production: a relatively fast aging
country becomes capital-abundant and labor-scarce vis-á-vis a slow aging country.
Under such circumstances, opening up a country for trade leads to the emergence of
demographically-induced Heckscher-Ohlin trade flows.
If demographic differences emerge as a potential determinant of international
trade flows, the question arises whether a country or a region can benefit from dif-
fusing part of its demographic shock abroad by means of liberalizing its trade policy.
Since trade creates a tendency to reduce the dispersion in international factor and
commodity prices, opening up an economy for trade might “cushion” the adverse
consequences of population aging. However, as real wages and interest rates move
in different directions the overall effect on income and welfare is not unambiguous.
Moreover, the dynamic paths of equilibrium factor prices hinge on the extent and
timing of the demographic change in both regions. Thus, the qualitative and quanti-
tative effects from factor price changes will be distinct for each generation. Lastly,
households are not only heterogeneous with respect to their age but also exhibit fun-
damental differences in preferences and ability. This point is taken into account by
allowing for 10 different household types within each generation.
46 3 Trade Liberalization and Global Demographic Change: A Quantitative Assessment
In contrast to what one would expect from standard neoclassical trade theory,
gains from trade in the presence of unsynchronized global aging patterns are not
guaranteed, i.e. openess may be immiserizing. This is a result of the evolution of
autarky-trade factor price differentials during the demographic transition. We find
that, for the fast aging region, trade liberalization is only beneficial for current old
generations that are born before and in the beginning of the demographic transition.
Future generations of all household types stand to incur substantial utility losses.
The intra-generational distribution of welfare gains and losses depends on the level
of labor productivity and the propensity to save throughout the life cycle. “Poor”
households, i.e. low-skilled and asset-poor households, tend to be relatively worse
off. We perform a systematic sensitivity analysis over a broad range of empirically
plausible parameters and find that there is a 80, 7% probability for households in the
fast aging region not to gain from trade liberalization. Moreover, looking at the bene-
fits of trade from a country/region perspective, we find that aggregate social welfare
gains require unrealistically high discount rates of about 8% and more. Despite of
the lack of social desirability for an open trade regime, the political assessment of
trade liberalization by means of an “hypothetical democratic referendum” under-
taken at the beginning of the demographic transition would support a liberal stance
of trade policy at the expense of future generations.
A substantial and growing literature has been calling attention to the economic
implications of unsynchronized global aging patterns in open economies. Besides
empirical studies2 computable overlapping generations (OLG) models have emerged
as the dominant analytical tool. For instance, Attanasio and Violante (2005), Feroli
(2003), Brooks (2003), and Domeij and Floden (2004) overwhelmingly find that
unequal demographic trends across countries can potentially induce global capital
flows if capital markets are integrated.3 A second strand of the literature analyzes the
role of pension reform and associated savings patterns for international capital flows
in the light of population aging (Attanasio and Violante (2005), Börsch-Supan et al.
(2006), INGENUE (2001)) and finds that capital flows from fast aging countries to
the rest of the world will initially be substantial but that trends are reversed when
aging progresses and households decumulate savings. Moreover, the status quo of
public pension systems is shown to crucially determine the magnitudes of interna-
2 See Bryant and McKibbin (1998, 2003) for good surveys of multi-country macroeconometric
simulation models that incorporate demographic change.
3 For example, Feroli (2003) uses a multi-region OLG model that is calibrated to match demo-
graphic differences among the major industrialized countries over the past 50 years and finds that
demographic differences can explain not all but some of the observed long-term capital movements;
Brooks (2003) finds that retirement saving by aging baby boomers will raise the supply of capital
substantially above investment in both the European Union and North America, causing both re-
gions to export large amounts of capital to Latin America and other emerging markets in the years
ahead. Beyond 2010, however, baby boomers in the European Union and North America will dis-
save in retirement, causing both regions to become capital importers. This shift will be financed by
capital flows from Latin America and other emerging markets, while Africa will remain dependent
on foreign capital for the foreseeable future because of continued high population growth. Domeij
and Floden (2004) empirically test to what extent international capital flows predicted by the model
match historically observed current account positions and find that a small but significant fraction
of international capital movements can be explained by the simulation model.
3.2 Model Formulation 47
tional capital movements. A series of papers by Fehr, Jokisch and Kotlikoff (2003),
Fehr et al. (2004a), and (n.d.) develop closed- and open economy OLG models
that allow, among many other features, for joint mobility of capital and labor across
regions. For instance, Fehr et al. (2004b) find that even a significant expansion of
immigration, whether across all skill groups or among particular skill groups, will
do remarkably little to alter the major capital shortage, tax hikes, and reductions in
real wages that can be expected along the demographic transition.
The rest of this chapter is organized as follows. Section 3.2 sets out the ana-
lytical framework. Section 3.3 explains our calibration strategy and specifies the
region-specific demographic dynamics. Section 3.4 presents the main model results
looking at the economic implications of unsynchronized global aging patterns under
autarky and trade, and also discusses the inter- and intragenerational distribution of
gains from trade. Section 3.4.5 performs a number of sensitivity analyses to pro-
vide a comprehensive quantitative assessment of trade liberalization under global
demographic change. Section 3.5 concludes.
4 In the international trade literature, OLG models are still used very rarely. Exceptions are models
that have been put forward by Mountford (1998), Sayan (2005), and Bajona and Kehoe (2006)—
all extending the seminal contribution by Galor (1992) to a two-country context. To maintain an-
alytical tractability, however, they abstract from a number of structural model components, e.g.,
multi-period lived agents and endogenous labor supply, that are pivotal for the type of analysis we
carry out here.
48 3 Trade Liberalization and Global Demographic Change: A Quantitative Assessment
that the model describes a world where a fully-funded pension system is in place that
implicitly operates through the (private) life-cycle savings behavior of households.
Ωh = {Πh , σh , νh , αh } (3.1)
where Πh denotes a labor productivity profile over the life cycle, σh is the intertem-
poral elasticity of substitution, σcl,h = 1/(1 − νh ) is the elasticity of substitution
between consumption and leisure, and αh determines the relative importance of ma-
terial consumption vis-à-vis leisure consumption. The presence of low and high-
skilled workers as well as different preferences with regard to the substitutability
of intra- and intertemporal consumption implies different savings and consumption
behavior over the life cycle that in turn give rise to qualitatively and quantitatively
different welfare implications of trade liberalization for households (for more details
on the empirical specification of household parameters see section 3.3.1). Note that
this setup implies that at each point in time there are H ×(N +1) different household
types that coexist. In the numerical examples below, we set N = 14 and H = 10, so
that H × (N + 1) = 150.
In each period over the life cycle households are endowed with units of time that
they allocate between labor and leisure. Households are assumed to be for-ward-
looking individuals that form rational point expectations (perfect foresight) over
the infinite horizon. Lifetime utility of generation g and type h in region r, ug,h,r ,
is additively separable over time and is of the constant-intertemporal-elasticity-of-
substitution form (CIES). The representative agent of each generation and type
chooses optimal consumption and leisure paths over his life cycle subject to lifetime
budget and time endowment constraints. The optimization problem for generation g
and household type h in region r is given by:
g+N g+N
∑ pr,a,t cr,g,h,t ≤ pr,k,t kr,g,h,g + ∑ pr,l,t πg,h,t (ωr,g − `r,g,h,t )
t=g t=g
`r,g,h,t ≤ ωr,g
cr,g,h,t ≥ 0 , `r,g,h,t ≥ 0 . (3.2)
where γr,g is the region-specific and time-dependent fertility rate. The size of the
generation born at the beginning of year zero is normalized to unity. Note that there
is no growth in time endowments over the life cycle. Thus, while the number of
households across generations increases over time, the size of a cohort over its life
cycle remains constant. Total population Nr,t in region r at time t is given by Nr,t =
H ∑tg=t−N ωr,g . The population growth rate γer,t can thus be defined as:
Nr,t+1
γer,t = −1. (3.4)
Nr,t
To maintain a balanced growth path, the age structure of the population has to be
stationary which requires that γ = γe = const.7
5 Households first decide how to allocate their lifetime income over time. Given the expenditure
for z, households decide in a second stage how much to spend on consumption and leisure. The
assumption of multi-stage budgeting is innocuous if and only if the utility function u is weakly
separable and the sub-utility functions z are homothetic. Both conditions are satisfied in this model.
6 Note that due to the convex structure of CES-preferences the nonnegativity constraints on c and
the working-age population ratio, denoted by Wr,t , which we define here as the number of people
50 3 Trade Liberalization and Global Demographic Change: A Quantitative Assessment
Production occurs within a period and technologies are stationary over time.8 Com-
petitive firms in both regions have access to two linearly homogenous produc-
tion technologies of the constant-elasticity-of-substitution (CES) form, index-ed by
i, j ∈ {1, 2}, that combine services of capital, Kt,r,i , and labor, Lt,r,i to produce a
homogenous good:
1/εi
εi εi
Yi,r,t = αi Ki,r,t + (1 − αi ) Li,r,t (3.6)
α1 > α2 . (3.7)
Capital intensity reversals at all factor and commodity prices are ruled out by the
functional form of (3.6) and the additional assumption that ε1 = ε2 .9 This assump-
tion is central to the results and determines along with changes in the relative abun-
dance of factors the structure of trade patterns.
The sectoral output good Yi,r,t can be traded on international markets. There are
no transportation costs or any other type of trade barriers. Domestically produced
and imported quantities of Yi,r,t are combined by a CES-aggregator to form a final
good At,r :
1/εa
Ar,t = αa aε1,r,t
a
+ (1 − αa ) aε2,r,t
a
(3.8)
H ∑tg=t−N+0.6 N ωr,g
Wr,t = . (3.5)
Nr,t
8 We solely focus on the role of population aging in determining growth dynamics and do not
consider technological change. Adding another exogenous engine of growth does not change the
results with respect to the implications of differential population dynamics.
9 See Wong (1990).
3.2 Model Formulation 51
In the following, let upper case letters denote the respective aggregate variable.
Given intergenerational heterogeneity, we impose the following feasibility condi-
tions to ensure that individual household behavior is consistent with the aggregate
economy:
t H
Lr,t = ∑ ∑ ωr,g,t − `r,g,h,t (3.9)
g=t−N h=1
t H
Ir,t = ∑ ∑ ir,g,h,t (3.10)
g=t−N h=1
t H
Kr,t = ∑ ∑ kr,g,h,t (3.11)
g=t−N h=1
t H
Cr,t = ∑ ∑ cr,g,h,t . (3.12)
g=t−N h=1
The law of motion for the aggregate capital stock for each region is given by:
In each period t, the composite consumption good can be used for consumption and
investment:
Ar,t = Cr,t + Ir,t . (3.17)
Finally, and to close the model, we impose the restriction that international borrow-
ing and lending is not permitted.10 In addition, we assume that neither of the regions
has fiat money, i.e. ∑0g=−N kr,g,h,g = (1 + r̄) K r . Here, r̄ and K r denote the steady state
interest rate and the base year capital stock in region r, respectively. The assumption
that international trade is balanced in each period can then be written as:
10 Bajona and Kehoe (2006) show in a similar setting that if international borrowing and lending
takes place, i.e. a country can run a temporary trade imbalance, the equilibrium patterns of capital,
international borrowing and lending, and trade are indeterminate.
52 3 Trade Liberalization and Global Demographic Change: A Quantitative Assessment
This condition can be derived by rewriting the lifetime budget constraint in (3.2) as
a series of periodic budget constraints, aggregating the periodic constraints over all
generations and household types that are alive at a given point in time, and finally
making use of Euler’ theorem for homogeneous functions.
3.2.4 Equilibrium
∞
1. Given prices {pr,a,t , pr,l,t , pr,re,t , pr,k,t , pi,t }t=0 , the choices {cr,g,h,t , `r,g,h,t }
solve the respective optimization problem in (3.2).
∞
2. Given prices {pr,a,t , pr,l,t , pr,re,t , pr,k,t , pi,t }t=0 , the representative firm in each
traded industry, i = 1, 2, and in the final goods sector maximizes profits, i.e.
{Ki,r,t , Li,r,t } ∈ arg maxKi,r,t ,Li,r,t ≥0 pi,t Yr,i,t (Ki,r,t , Li,r,t ) − pr,re,t Ki,r,t − pr,l,t Li,r,t
and
{a1,r,t , a2,r,t } ∈ arg maxa1,r,t ,a2,r,t ≥0 pr,a,t Ar,a,t (a1,r,t , a2,r,t ) − p1,t a1,r,t − p2,t a2,r,t ,
respectively.
∞
3. The sequences of household choices {cr,g,h,t , `r,g,h,t }t=0 , and production
∞ ∞
plans, {Yi,r,t , Ki,r,t , Li,r,t }t=0 , and {Ar,t ,Y1,r,t ,Y2,r,t }t=0 , satisfy the feasibility con-
ditions (4.33)–(3.18).
n
Definition 2: A steady state is a collection of choices for households cbr,g,h,t , `br,g,h,t
∞ ∞
t=0 , and for the representative firm in each traded industry {Yi,r,t , Ki,r,t , Li,r,t }t=0 ,
b b b
and for the representative firm in the final goods sector, {A br,t , Yb1,r,t , Yb2,r,t }∞ , and
∞
t=0
prices pr,a,t , pr,l,t , pr,re,t , pr,k,t , pi,t }t=0 , that satisfy the conditions of a competitive
equilibrium for appropriate initial endowments of capital kr,g,h,g = b kr,g,h,g , and where
the age structure of the population is stationary, i.e. γ = γe = const. Here, we set
νt = νb for all t, where ν represents a generic variable.
3.3 Calibration and the Nature of Demographic Shocks 53
The model outlined above does not possess an analytical solution and therefore has
to be solved numerically. Appendix 3.5 formulates the dynamic general equilibrium
of the model as a mixed-complementary problem (MCP) and provides more details
on the steady state calibration of the model and the solution algorithm that is used
to compute the transitional dynamics and steady state effects of the demographic
simulation experiments. For solving the model, we use the decomposition algorithm
presented in Chapter 2. In the present context, the equilibrium of the OLG economy
is approximated by solving a sequence of appropriately chosen two-country, two-
sector Ramsey growth models.
Given the oversimplified and illustrative nature of the model, we adopt a naive cali-
bration approach that builds on standard values from the literature.11 Section 3.4.5
performs a number of sensitivity analyses to check for the robustness of the results.
For the technology parameters, we set α1 = 0.7, α2 = 0.3, 1/(1 − εi ) = 1, i = 1, 2,
αa = 0.5, 1/(1 − εa ) = 1, and δ = 0.1.12 The model is calibrated to a steady state
equilibrium where we assume that along the baseline path quantities grow with
1 + γ̄ = 1.02 and present value prices decline with 1 + r̄ = 1.1. We reconcile in-
dividual decisions by the OLG households with aggregate data in the base year by
11 Numerical values for the exogenous model parameters are taken from Backus, Kehoe and Kyd-
land (1992) and Ambler, Cardia and Zimmermann (2002).
12 For the base case, we assume Cobb-Douglas technologies for Y
i,r,t and Ar,t , i.e. εi = εa = 0. For
these values the CES functions in (3.6) and (3.8) are not defined. However, as εi and εA approach
zero, the isoquants of the CES production functions look very much like the isoquants of the Cobb-
Douglas production function.
54 3 Trade Liberalization and Global Demographic Change: A Quantitative Assessment
6 c o n s u m p tio n
la b o r in c o m e p r o d u c tiv ity in d e x
1 0 0
c a p ita l in c o m e w o r k in g tim e ( in % )
le is u r e tim e ( in % )
in d e x ( fir s t p e r io d c o n s u m p tio n = 1 )
6 0
4 0
2
2 0
0 0
0 5 1 0 1 5 0 5 1 0 1 5
A g e in life c y c le ( b e g in n in g o f p e r io d ) A g e in life c y c le ( b e g in n in g o f p e r io d )
(a) Baseline profiles and asset stocks (b) Productivity index and allocation of time
Fig. 3.2 Calibrated life cycle behavior for h = 3 (initial steady state)
Note: Optimal life cycle behavior for all generations and household types along the baseline path is
derived in the first calibration step using a steady state assumption. For details see Appendix 3.5.1.
The household calibration model is parameterized as shown in Table 3.1.
calibrating the discount factor ρ such that the sum of the value of individual con-
sumption demands equals aggregate consumption. Likewise, to match individual
asset holdings with the value of the aggregate capital stock in the base year, we in-
troduce a scaling factor on the time endowment of households and select its value
accordingly.13
Naturally, intra-period and intertemporal household behavior crucially hinges on
assumptions regarding various elasticities parameters. The way of parameterizing
household behavior takes the extreme standpoint that the uncertainty of empirical
parameter estimates is to a large extent a result of differences in underlying house-
hold preferences. Hence, we incorporate “uncertainty” in a direct fashion by allow-
ing for different preferences within each age group. In 3.4.5 we use Monte-Carlo
simulation methods to check for the general robustness of results over a broad range
of empirically plausible parameter configurations.
Following Auerbach and Kotlikoff (1987), we assume that labor productivity fol-
lows a hump-shaped profile over the life cycle:
2
πa,h = ζscale,h e(4.47 + ζage,h a − ζage2,h a ) / e4.47 . (3.19)
For each household type, we slightly perturb their numerical specification thereby
allowing for differences in labor productivity. Labor productivity across generations
of a given household type is assumed to be identical. Table 3.1 shows the parameter
specification which is used in the numerical simulation later on. The model is solved
in one-year intervals and N = 14 which implies that households live for 15 years.
13 Since income is linear in this scaling factor and demand is linear in income, such a scaling
parameter has no economic significance. For more details of the steady state calibration procedure
see Appendix 3.5.1.
3.3 Calibration and the Nature of Demographic Shocks 55
2 ,0 H o u s e h o ld ty p e :
1 (p o o re s t)
2
o f e c o n o m y - w id e a s s e ts
1 ,5 3
4
5
6
1 ,0 7
8
9
1 0 ( r ic h e s t)
0 ,5
%
0 ,0
0 5 1 0 1 5
A g e in life c y c le ( b e g in n in g o f p e r io d )
Note: The plot shows the amount of assets held by different household types through the life cycle
as a percentage fraction of economy-wide assets (along the steady state baseline path).
This is found to be sufficient for generating a meaningful life cycle behavior on the
part of the households and for studying the implications of population aging.
Figure 3.2, Panel (a), shows the calibrated steady state income and consumption
profiles and asset stocks over the life cycle for a household of type h = 3. For the in-
terpretation of the results later on, it is important to understand the nature of the life
cycle behavior which builds on the foundation developed by Modigliani and Brum-
berg (1954). Households arrange savings so as to smooth consumption over their
lifetime—as it is implied by the intertemporal Euler equation. Due to the hump-
shaped productivity profile, time devoted to labor and consequently labor income is
increasing in the first years and then decreases—as it is implied by the intratempo-
ral Euler equation for leisure. Households accumulate assets in younger to middle
ages and run down their asset positions when being old; capital income evolves ac-
cordingly. In the last period of the life cycle, consumption is entirely financed out
of savings since households optimally chose not to earn labor income. Figure 3.2,
Panel (b), shows the allocation of leisure and working time over the life cycle. With
the given parametrization, labor supply drops to zero in the two last periods of the
life cycle with the reservation wage exceeding the market wage. An attractive fea-
ture of this analysis is the use of complementarity programming to accommodate
endogenous regime shifts—as here, for instance, optimal retirement decisions by
households.
As a consequence of the differences in intra- and intertemporal behavior of house-
holds, there are large dissimilarities with respect to the optimal wealth accumulation
path. Figure 3.3 illustrates this point and shows the amount of assets that is being
held by each household type throughout the lifetime as a fraction of economy-wide
assets.
56 3 Trade Liberalization and Global Demographic Change: A Quantitative Assessment
2 ,0
fa s t a g in g r e g io n 7 2 fa s t a g in g r e g io n
s lo w a g in g r e g io n s lo w a g in g r e g io n
W o r k in g - a g e p o p u la tio n r a tio ( in % )
P o p u la tio n g r o w th r a te ( in % )
1 ,5 7 1
1 ,0
7 0
0 ,5
6 9
0 ,0
0 5 1 0 1 5 2 0 2 5 3 0 3 5 0 5 1 0 1 5 2 0 2 5 3 0 3 5
T im e p e r io d T im e p e r io d
Note: Panel (a) shows the changes in the population growth rate γer,t as implied by the fertility rate
shocks. Panel (b) shows the impact on the working-age population ratio Wr,t .
Population aging is modeled by assuming that fertility rates decrease over time.14
Shocks to the fertility rate rate alter the age composition of the population. Globally
unsynchronized aging patterns are modeled by assuming that the speed of decline in
fertility rates varies across regions. Region a is taken to be the fast aging region vis-
à-vis a slow aging region b. In the initial steady state, both regions are characterized
by an identical fertility rate γ̄. From year zero onwards population dynamics begin
to diverge and both regions experience a linear decline in fertility rates. Eventually,
fertility rates in both region converge to a new and common long-run level γ SS < γ̄.
More specifically, the time profile for the fertility rates in region r is given by:
γ̄ if g ≤ 0
γr,g = γ̄ − ∆r t if 0 < g ≤ Br (3.20)
SS
γ if g > Br ,
14 The number of periods over the life cycle is kept constant. Hence, we do not consider changes
in longevity.
3.4 Model Results and Discussion 57
15 Although this model features elastic labor supply on the part of the households, we do not
distinguish between the “per capita” and “per worker” concept. From a qualitative viewpoint both
concepts lead to identical results while quantitative differences are negligible.
16 The result of a higher steady state level of per capita capital and output in response to a perma-
nently lower population growth rate is also obtained in a Solow-Swan growth framework. In the
Ramsey model, however, changes in the population growth rate have only transitional effects on
these per capita variables since there is by construction no “leverage” for the compositional shift
in the age structure of a population. For more details see, e.g., Barro and Sala-i-Martin (2004).
58 3 Trade Liberalization and Global Demographic Change: A Quantitative Assessment
1 2 4
8
s te a d y s ta te ( in % )
s te a d y s ta te ( in % )
2
re a l w a g e ra te
4
c a p ita l p u r c h a s e r a te
r e a l c a p ita l r e n ta l r a te
0
0
c o n s u m p tio n
in v e s tm e n t
-4
D e v ia tio n fr o m
D e v ia tio n fr o m
c a p ita l s to c k
o u tp u t -2
-8
-1 2
-4
0 1 0 2 0 3 0 4 0 0 1 0 2 0 3 0 4 0
T im e p e r io d T im e p e r io d
(a) Per capita variables, fast aging region (b) Real factor prices, fast aging region
1 ,0 7
c o n s u m p tio n
in v e s tm e n t
6
c a p ita l
s te a d y s ta te ( in % )
s te a d y s ta te ( in % )
0 ,5 o u tp u t
5
4
0 ,0
3
fa s t a g in g r e g io n
D e v ia tio n fr o m
D e v ia tio n fr o m
-0 ,5 s lo w a g in g r e g io n
2
1
-1 ,0
0
0 1 0 2 0 3 0 4 0 0 1 0 2 0 3 0 4 0
T im e p e r io d T im e p e r io d
(c) Per capita growth rates, fast aging region (d) Per capita capital stock, fast aging region
Note: The plots show the transition to a new long-run equilibrium in response to demographic
shocks as specified in eq. (3.20). Panel (a): Consumption is defined as total private sector consump-
tion. Investment is gross investment. Output is defined as the quantity of the composite good A
produced at a given point in time. Panel (b): Factor prices are nominal factor prices deflated by the
consumer price index.
capital stock and the fact that output has to be shared among less people. Finally,
since in this model population growth is the only engine of long-run growth, growth
rates of per capita variables converge back to zero in the new steady state equilib-
rium (Panel (c)). Section 3.4.5 demonstrates that these outcomes are qualitatively
robust over a broad range of parameter configurations that is supported by empirical
literature.
Factor price changes imply different costs of production for each sector and there-
fore bring about a sectoral change in the aging economy. Figure 3.6 shows that the
output of the capital-intensive sector expands relative to the labor-intensive sector
(Panel (a)) and that the relative price of the capital-intensive commodity declines
(Panel (b)). The movement in factor price changes is an instance of what is known
as the “Rybczynski effect”: the fall in the relative price of commodity 1 lowers the
real return to the factor used intensively in the production of that good, here capi-
tal, and hence the capital rental rate falls, and increases the real return to the other
3.4 Model Results and Discussion 59
factor, here labor, and hence the wage rate rises.17 Since in each period changes
in commodity prices are a weighted average of changes in factor prices, the wage
rate (capital rental rate) increases (decreases) in percentage terms by more than the
relative price of commodity 1.
We can summarize the key results of this section as follows:
1. Population aging leads to a higher per capita capital stock during the transition
and in the new long-run equilibrium.
2. In a two-sector economy in which sectoral output is produced with different fac-
tor intensities, population aging leads to a relative contraction (expansion) of the
labor-intensive (capital-intensive) sector. The real wage rate (capital rental rate)
increases (decreases) during the transition and in the new long-run equilibrium.
Differences in the extent and timing of aging patterns across regions lead to quanti-
tatively distinct responses in both economies. Figure 3.5, Panel (d), shows that the
capital deepening process is less pronounced in the slow aging region as compared
to the fast aging region. Since the working-age population ratio in the fast aging re-
gion is higher throughout the demographic (and economic) transition, the per capita
capital stock in the slow aging region is lower than in the fast aging region, i.e. the
slow (fast) aging region becomes relatively labor (capital) abundant. As a result,
the ”dampened Rybczynski effect is weaker for the slow aging region implying a
weaker relative expansion of the capital-intensive sector (see Figure 3.6, Panel (a)).
Under autarky, differences in relative factor endowments cause equilibrium fac-
tor prices to diverge (Figure 3.6). The wage rate (capital rental rate) in the capital-
abundant/labor-scarce region is higher (lower) than in the capital-scarce, labor-
abundant region. Despite identical production technologies in both regions, differ-
entials in factor prices imply different costs of production which in turn translate
into international relative price differences of sectoral output under autarky (Figure
3.6, Panel (b)). The fast aging region therefore has a cost advantage in producing the
capital-intensive commodity whereas the slow aging region has a cost advantage in
the production of the labor-intensive commodity. Hence, the relative price of com-
modity 1 in the fast aging region stays below the one in the slow aging region as
long as differences in relative factor endowments across regions are sustained.
Divergence of relative commodity prices under autarky create incentives for
trade. If international trade is liberalized, each region will export the good where
it has a cost advantage. Exploiting the higher relative price of commodity 1 in the
slow aging region, the fast aging, capital-abundant region consequently exports the
17Note that we only observe a ”dampened” Rybczynski effect due to feedback effects of induced
commodity price changes which cushion—but not reverse—the ”pure” Rybczynski effect that
would be obtained if prices were fixed. See, e.g., Jones (1965, pp.562).
60 3 Trade Liberalization and Global Demographic Change: A Quantitative Assessment
0
a u ta r k y , fa s t a g in g r e g io n
a u ta r k y , s lo w a g in g r e g io n
tra d e
s te a d y s ta te ( in % )
s te a d y s ta te ( in % )
2
-1
1
D e v ia tio n fr o m
D e v ia tio n fr o m
a u ta r k y , fa s t a g in g r e g io n
a u ta r k y , s lo w a g in g r e g io n -2
0
0 1 0 2 0 3 0 4 0 0 1 0 2 0 3 0 4 0 5 0
T im e p e r io d T im e p e r io d
0
3 a u ta r k y , fa s t a g in g r e g io n
a u ta r k y , s lo w a g in g r e g io n
tra d e
s te a d y s ta te ( in % )
s te a d y s ta te ( in % )
-1
2
D e v ia tio n fr o m
D e v ia tio n fr o m
-2
1 a u ta r k y , fa s t a g in g r e g io n
a u ta r k y , s lo w a g in g r e g io n
tra d e
0 -3
0 1 0 2 0 3 0 4 0 5 0 0 1 0 2 0 3 0 4 0 5 0
T im e p e r io d T im e p e r io d
Fig. 3.6 Sectoral change and impact on equilibrium commodity and factor prices
Note: The plots show the transition to a new long-run equilibrium in response to demographic
shocks as specified in eq. (3.20). The relative size of the capital-intensive sector 1 is measured as
the ratio of sectoral outputs. The relative price of commodity 1 is defined as the ratio of the price
for commodity 1 and 2. Real factor prices are nominal factor prices deflated by the consumer price
index.
ital in the labor abundant region, which exports the labor-intensive good, is higher
under trade as compared to autarky (not shown).
We can summarize the key results of this section as follows:
1. Globally unsynchronized aging patterns affect the relative abundance of factors
of production across regions and lead to differentials in commodity and factor
prices under autarky. The per capita capital stock in the fast aging region is higher
than in the slow aging region.
2. If international goods markets are integrated (and if goods are produced with dif-
ferent factor intensities), the presence of globally unsynchronized aging patterns
leads to demographically-induced Heckscher-Ohlin trade. The fast (slow) aging
region specializes in the production and export of the capital-intensive (labor-
intensive) commodity during the transition.
In contrast to what one would expect from standard neoclassical trade theory, mov-
ing from an autarky situation to a world where goods can move freely across bor-
ders does not unambiguously result in welfare gains on a household level. Depend-
ing on the particular set of socio-economic characteristics of a household type and
generation—including age, labor productivity and other principal characteristics
that determine intra- and intertemporal behavior—gains from trade are not guaran-
teed. Figure 3.7 shows the percentage change in Hicksian equivalent variation (EV)
by generation and household type for the fast aging region.18
On an intergenerational basis, trade liberalization in the presence of globally
unsynchronized aging pattern has very distinct quantitative and qualitative welfare
consequences for current and future generations. Depending on the specific house-
hold type, generations born in the run up to and the beginning of the demographic
transition gain (lifetime utility of generations g = −15, . . . , 8 increases) whereas gen-
erations born in the midst of the transition and thereafter (generations g > 9 for all
household types) stand to incur substantial utility losses. As the age structure of
populations in both regions converges, the adverse welfare effects for newly born
generations diminish.
On an intragenerational basis, whether welfare gains materialize for current old
generations depends on the particular household characteristics. Figure 3.8, Panel
(a), takes a cross-sectional view and shows the distribution of welfare changes across
household types for generation g = 5. It is evident that “poor” households, i.e. low-
skilled workers and those households that hold relatively few assets throughout their
life cycle, lose from trade liberalization whereas richer households gain overall.
18 Due to the artificial symmetry of this two-country model, welfare changes for the slow aging
region are diametrically opposed to what is observed in the fast aging region. In the following
discussion, we focus on the fast aging region.
62 3 Trade Liberalization and Global Demographic Change: A Quantitative Assessment
0 ,8
H o u s e h o ld ty p e :
C h a n g e in H ic k s ia n e q u iv a le n t v a r ia tio n ( in % )
1 (p o o re s t)
0 ,4 2
3
4
0 ,0 5
6
7
-0 ,4
8
9
1 0 ( r ic h e s t)
-0 ,8
-1 ,2
-1 5 -5 5 1 5 2 5 3 5 4 5 5 5
G e n e r a tio n
Note: Changes in Hicksian equivalent variation resulting from a trade liberalization in the presence
of globally unsynchronized aging patterns. The graph shows welfare changes by generation and
household type for the fast aging region.
0 ,3
C h a n g e in H ic k s ia n e q u iv a le n t v a r ia tio n ( in % )
0 ,2
0 ,1
0 ,0
1 2 3 4 5 6 7 8 9 1 0
-0 ,1
H o u s e h o ld ty p e
-0 ,2
-0 ,3
-0 ,4
-0 ,5
Note: Changes in Hicksian equivalent variation resulting from a trade liberalization in the presence
of globally unsynchronized aging patterns across household types for generation g = 5, fast aging
region.
Why are there winners and losers from trade liberalization in the presence of glob-
ally unsynchronized aging patterns ? What explains that future generations will ex-
perience welfare losses whereas current old generations gain ? Why are low-skilled
workers and those households that exhibit relatively low levels of savings throughout
their lifetime most adversely affected by global demographic change ? The answers
lie in the evolution of autarky-trade factor price differentials. The pressure on real
factor prices that stems from the region-specific demographic change is mitigated
by the specialization patterns in production and the international exchange of goods.
3.4 Model Results and Discussion 63
G e n e r a tio n
0 ,5 0 (ty p e h = 1 ):
0 ,4 re a l w a g e ra te g = - 1 4
C h a n g e in m a te r ia l c o n s u m p tio n ( in % )
r e a l c a p ita l r e n ta l r a te g = - 1 3
g = - 1 0
0 ,2 5 g = - 8
g = - 6
a u ta r k y ( in % )
0 ,2
g = - 4
g = - 2
0 ,0 0
g = 0
g = 2
0 ,0
g = 4
-0 ,2 5 g = 6
C h a n g e fro m
g = 8
g = 1 0
-0 ,2 g = 1 2
-0 ,5 0 g = 1 4
g = 1 6
g = 1 8
-0 ,4 g = 2 0
-0 ,7 5
0 1 0 2 0 3 0 4 0 5 0 0 5 1 0 1 5 2 0 2 5 3 0 3 5
T im e p e r io d T im e p e r io d
G e n e r a tio n 2 0 G e n e r a tio n
(ty p e h = 1 ): (ty p e h = 6 ):
0
g = - 1 4 g = - 1 4
C h a n g e in la b o r s u p p ly ( in % )
g = - 1 3 g = - 1 3
C h a n g e in la b o r s u p p ly ( in % )
1 5
g = - 1 0 g = - 1 0
-1
g = - 8 g = - 8
g = - 6 g = - 6
g = - 4 1 0 g = - 4
-2 g = - 2 g = - 2
g = 0 g = 0
g = 2 g = 2
-3 g = 4 5 g = 4
g = 6 g = 6
g = 8 g = 8
g = 1 0 g = 1 0
-4 g = 1 2 0 g = 1 2
g = 1 4 g = 1 4
-5
0 5 1 0 1 5 2 0 2 5 3 0 0 5 1 0 1 5 2 0 2 5 3 0
T im e p e r io d T im e p e r io d
Fig. 3.9 Factor price changes and impact on household decisions (trade vs. autarky)
Note: The plots show compare trade with autarky under population dynamics as specified in (3.20).
All plots refer to the fast aging region. Panel (a): Change in real factor prices. Panel (b): Change in
material consumption, cg,h,t , for household type h = 1. Panel (c) and (d): Change in labor supply,
ωg,t − `g,h,t , for household type h = 1 and h = 6, respectively.
International trade equalizes the real factor prices and therefore increases the real
capital rental rate and decreases the real wage rate in the fast aging region relative
to autarky (Figure 3.9, Panel (a)).
International trade can thus be viewed as a mechanism which arbitrages away
differences in global demographic patterns as the integrated world economy as a
whole now faces an “average demographic scenario” with factor and commodity
prices being averages of autarky equilibrium prices. Changes in real factors prices
alter the consumption and labor supply/leisure choices of households directly affect-
ing utility as in (3.2) and therefore explain why some households gain and others
lose.
Consumption and labor supply decisions are equilibrium outcomes that de-
pend on a number of direct and indirect effects—some of them work in different
directions—and that are determined by key household parameters and by the mag-
nitude of prices changes that households experience over their lifetime.19 Consump-
tion choices over the life cycle of a generation are directly affected by changes in
the real interest rate since, firstly, for a given level of savings higher interest rates
make households wealthier (wealth effect) and, secondly, changes in the interest rate
affect the relative price of intertemporal consumption. Specifically, higher interest
rates imply that consumption today becomes more expensive vis-à-vis consumption
tomorrow, and hence households allocate consumption towards later periods of their
life (substitution effect). Second, there is an indirect effect on consumption which
stems from changes in the wage rate: for a given level of labor supply, lower wage
rates depress lifetime income and consumption therefore decreases (wage effect).
Labor supply decisions respond to changes in the real wage rate (price effect). The
intuitive case is that labor supply is increasing in the wage rate. Intertemporally op-
timizing households may, however, increase labor supply in response to a decline
in the wage rate so as to prevent large fluctuations in periodic income which would
be inconsistent with the objective of consumption smoothing. In such a case, leisure
consumption is a Giffen good. Interest rate changes also affect labor supply deci-
sions since for a given level of savings, higher interest rates imply that households
have become more wealthy, and because there is disutility from working this creates
a tendency for labor supply to fall (interest rate effect).
Figure 3.9, Panel (b), shows the change in material consumption, cg,h,t , over the
respective lifetime for each generation g = −14, . . . , 20 of household type h = 1
in the fast aging region. Recall from Panel (a) that the gap between autarky and
trade factor prices increases during the demographic transition and diminishes even-
tually as the working-age population ratios in both region converge. For generations
born prior to year zero, we see that old-age consumption is slightly increased due
to a positive wealth and substitution effect. Consumption in earlier periods is al-
most unchanged because the drop in wage rates and hence negative wage effects
are relatively small. For subsequent generations, i.e. moving along the x-axis from
the left to the right in Panel (b), losses in wage income become increasingly sub-
stantial and lead to larger and larger reductions of consumption in earlier periods of
the life cycle. At the same time, the interest rate gap widens which strengthens the
positive income and intertemporal substitution effect causing increases in old-age
consumption. However, as the extent of real factor price changes over the lifetime
of a generation becomes less favorable, losses in wage income cannot be compen-
sated for by the positive income effect and hence there is a downward adjustment
in consumption for all periods of the life cycle (g ≥ 16). From year 17 onwards, the
trade-autarky differentials in factor prices begin to diminish and households born
during this period experience smaller and smaller decreases in wage rates over their
lifetime. Hence, moving across generations, the size of reductions in young-age con-
19 For the interpretation of results, we concentrate on the link from price changes to household
quantity decision, and neglect the fact that in general equilibrium prices and quantities are deter-
mined simultaneously.
3.4 Model Results and Discussion 65
0,10
0,05
-0,05
η=1
η = 0.8
-0,10
η = 0.6
η = 0.4
-0,15 η = 0.2
-0,20
0% 2% 4% 6% 8% 10% 12% 14%
Note: Change in the social welfare as defined in (3.21) comparing trade with autarky for different
values of ρe and Σ (fast aging region). A low value of ρe on the x-axis implies a high weight on
the utility of future generations. If η = 1, a more utilitarian approach is adopted, whereas smaller
values of η mean that a greater weight is placed on equity.
In this section, we take a first stab on whether trade liberalization in a world which
is characterized by differential population dynamics is desirable from a social stand-
point. We apply a very direct social welfare function approach and specifically as-
sume that aggregate welfare can be measured as:
!1/η
SWFr = ∑ Θr,g,h uηr,g,h (3.21)
g,h
The critical weakness of the concept of a social welfare function is the more or
less arbitrary choice on how to trade off utility between current and future genera-
3.4 Model Results and Discussion 67
tions. Figure 3.10 therefore shows the sensitivity of aggregate welfare changes in
the fast aging region with respect to the social discount rate and the interhousehold
substitution elasticity. A clear picture emerges: trade liberalization in the presence
of globally unsynchronized aging patterns is only beneficial for unrealistically high
social discount rates of about 8% and greater.20
Another perspective on the social desirability of trade liberalization in a world of
global demographic change can be provided by conducting an “hypothetical demo-
cratic referendum”. As in a stylized one-person, one-vote Western democracy, we
assume that each generation and household type alive at a given point in time can
dispense one vote which carries equal weight. A vote for trade liberalization is taken
if the change in the Hicksian equivalent variation is positive; otherwise, people
vote against it. In such a referendum, trade liberalization would almost surely be
voted down—except for the first periods of the demographic transition (up to year
6) where current old generations who benefit from trade are alive. fThe political
outcome of this thought experiment, however, is quite stark: although trade liber-
alization has been shown not to be beneficial from a social perspective, prevailing
political preferences at the beginning of the demographic transition would support
a liberal stance of trade policy at the expense of future generations.21
We can summarize the key results of this section as follows:
1. From a social standpoint, trade liberalization for the fast aging region in the pres-
ence of globally unsynchronized aging patterns is only beneficial for a range of
unrealistically high social discount rates.
2. Although not beneficial from a social perspective, the political assessment of
trade liberalization by means of an “hypothetical democratic referendum” under-
taken at the beginning of the demographic transition, would support a liberal
stance of trade policy at the expense of future generations.
In order to explore the robustness of the results with respect to key model parame-
ters, and to develop a better understanding of the impact of each single parameter
in the model, we first conduct a number of “piecemeal” sensitivity analyses. Table
3.2 shows results from variations in either δ , εi , i = 1, 2, or εa while holding all
other parameters constant. The figures in the table refer to percentage deviations of
20 As a first approximation, it seems reasonable to use the market interest rate for discounting social
investments, thereby respecting private preferences. Also, it is often argued that policy makers are
more patient than private citizens, therefore suggesting a social discount rate below the interest rate
(Caplin and Leahy, 2004).
21 Of course, there is a time inconsistency problem here. Future generations want current old
generation not to liberalize trade, yet if international goods markets were to remain separated, they
themselves would vote for trade liberalization.
68 3 Trade Liberalization and Global Demographic Change: A Quantitative Assessment
Note: The table shows results from “piecemeal” sensitivity analysis with respect to δ , εi , i = 1, 2,
and εa (only one parameter is changed while the others are being held constant). The figures refer to
deviations of steady state values of per capita variables in response to population aging as specified
in Section 3.3. All figures refer to the fast aging region. For the “piecemeal” sensitivity analysis all
household parameters are held constant.
steady state values in response to population aging in the fast aging region under the
autarky case. Varying key technology parameters confirms that the results arising
from solving the model with the base case parametrization are qualitatively robust.
In all cases, macroeconomic variables move in same directions. The intensity of eco-
nomic adjustment in response to the demographic shocks is the smaller the higher
is δ , and the lower are σY,i and σA . Intuitively, a higher rate of depreciation leads to
less capital accumulation. A higher elasticity of substitution between capital and la-
bor means that it is possible to increasingly substitute away from costly labor inputs
to capital services which have become relatively cheap due to the capital deepening.
This implies lower costs of production which in turn translate into smaller reductions
in investment, a higher capital stock, and stronger (positive) reactions of output and
consumption. Similarly, a higher substitutability between sectoral goods means that
the price of the final good A decreases as input costs decline due to a shift towards
the less expensive capital-intensive commodity. Furthermore, the piecemeal sensi-
tivity analysis reveals that the ranking in terms of the quantitative effects as shown
in Table 3.2 is also maintained during the transition phase (not shown).
3.4 Model Results and Discussion 69
0 ,0 6
0 ,0 5
0 ,0 4
D e n s ity
0 ,0 3
0 ,0 2
0 ,0 1
0 ,0 0
0 2 4 6 8 1 0 1 2 1 4 1 6
D e v ia tio n fr o m in itia l s te a d y s ta te ( in % )
Note: Sample frequency distribution of the percentage deviation of the per capita steady state capi-
tal stock from the initial steady state under autarky (fast aging region). In the systematic sensitivity
analysis, we execute 1,000 simulations of the autarky model drawing key household and technol-
ogy parameters from uniform probability distributions.
In order to get a comprehensive picture of what drives the results, and to allow for
the interaction of all parameters in the model, we conduct a systematic sensitivity
analysis and execute 1,000 simulations of, both, the autarky and trade model. In
each simulation, values for the key parameters are drawn from uniform probability
distributions. The support for the distribution of each parameter is chosen so as to
cover a range of empirically plausible parameter values that is well-supported by
the literature; we mainly draw on Backus et al. (1992) and Ambler et al. (2002).22
Figure 3.11 plots the sample frequency distribution of the percentage deviation of
the per capita steady state capital stock from the initial steady state under autarky
(fast aging region). The main message arising from this graph is that population
aging in this economy unambiguously leads to a positive impact on the per capita
capital stock. The sample mean of this distribution is an increase of 4.15%. There is
a 83.5% chance that the increase in the per capital capital stock is smaller than 7%.
Similarly, we perform a systematic sensitivity analysis to check for the sensitiv-
ity of welfare gains and losses from trade liberalization in the presence of globally
unsynchronized aging patterns (Figure 3.12). To compute the “overall’ sample fre-
22 The lower and upper bounds for the uniform probability distributions of each respective parame-
ter are: 0.5 < εi < 5, 0.5 < εa < 5, 0.1 < δ < 1, 0.5 < αh < 0.9, 0.25 < σh < 1.5, 0.25 < σcl,h < 1.5.
In all simulations, labor productivity is parameterized as in the base case scenario.
70 3 Trade Liberalization and Global Demographic Change: A Quantitative Assessment
A major concern of the sensitivity analysis was to check for the potential existence
of multiple equilibria. Galor (1992) pointed out the possibility of global indetermi-
nacy of the perfect-foresight equilibrium in the two-sector neoclassical OLG model.
In principle, this contingency cannot be ruled out by the specific structure and as-
sumptions of the model. The non-existence of a closed form solution makes it im-
possible to analytically characterize the dynamical system and establish conditions
for uniqueness. Results from the previous sensitivity analyses, however, suggest that
changes in key parameters of the model do not lead to qualitative reversals. More-
over, varying parameter values alters equilibrium outcomes of endogenous variables
in a ’monotonic way’. Finally, we experimented with the choice of T . Kehoe and
Levine (1985) have shown that in the OLG framework indeterminacy would mani-
fest itself as sensitivity to the truncation date. None of the models presented here are
sensitive to T , provided that it is sufficiently large. We interpret all these findings as
evidence that the equilibria are unique.
Despite the fact that the existing literature has consistently identified globally un-
synchronized aging patterns to be a potentially important driving force of cross-
border flows of capital and labor, the implications for the international exchange
of goods have been largely overlooked. This chapter fills the gap and develops a
numerical dynamic general equilibrium framework that combines elements of the
factor-proportions theory of international trade with those of Auerbach and Kot-
23 Running 1,000 simulations of the trade model drawing key parameters from uniform probability
distributions takes approximately six days on a dual core 2 GHz computer.
3.5 Concluding Remarks 71
0 ,8
0 ,7
0 ,6
0 ,5
D e n s ity
0 ,4
0 ,3
0 ,2
0 ,1
0 ,0
-7 -6 -5 -4 -3 -2 -1 0 1 2 3 4
C h a n g e in H ic k s ia n E V ( in % )
Fig. 3.12 Sample frequency distribution of welfare gains from trade liberalization
Note: Sample frequency distribution of the percentage change in Hicksian equivalent welfare com-
paring trade with autarky in the presence of globally unsynchronized aging patterns (fast aging
region). In the systematic sensitivity analysis, we execute 1,000 simulations of the autarky and
trade model drawing key household and technology parameters from uniform probability distri-
butions. The sample comprises in total 710, 000 households that are alive during the economic
transition phase and which differ with respect to their age, i.e. the point in life they are affected by
demographic shocks, and labor productivity and other key behavioral parameters that govern the
intra- and intertemporal behavior.
likoff (1987)-type OLG models to analyze the economic consequences of trade lib-
eralization in the presence of global demographic change.
We demonstrate that demographics emerge as a potential determinant of inter-
national trade flows. To the extent that age-dependent savings and labor supply
decisions by households have an impact on the aggregate capital stock and labor
force, population aging alters the relative abundance of factors of production and
thereby—in the presence of globally unsynchronized aging processes—gives rise to
Heckscher-Ohlin trade patterns. International trade can be viewed as a “mechanism”
which potentially arbitrages away regional demographic differences, and thereby
mitigates the pressure on real wages and interest rates that is created by population
aging in a closed economy.
Moreover, we provide a detailed analysis of the inter- and intragenerational dis-
tribution of gains from trade. In contrast to what one would expect from standard
neoclassical trade theory, gains from trade under regional unsynchronized aging
patterns are not guaranteed. Who gains from trade liberalization depends on their
specific set of socio-economic characteristics. We find, for instance, that in the fast
aging region current old generations born before and in the beginning of the de-
mographic transition benefit whereas future generations stand to incur substantial
72 3 Trade Liberalization and Global Demographic Change: A Quantitative Assessment
utility losses. The intragenerational distribution of welfare gains and losses depend
on the level of labor productivity and the propensity to save throughout the life cycle.
Low-skilled workers and asset-poor households tend to be relatively worse off. On
a regional level, gains from trade for the fast aging region require implausibly high
social discount rates. The results are robust over a broad range of empirically plausi-
ble parameter configurations. We perform a systematic sensitivity analysis and find
that there is 80.7% probability that households in the fast aging region do not gain
from trade.
Of course, the results obtained are subject to several caveats which may also
serve as suggestions for future research. First, by focussing on the goods market
channel as the only linkage between countries, the critical assumption is made that
factors of production are internationally immobile. Integrating cross-border migra-
tion and international capital flows into the model would add an additional mecha-
nism to arbitrage away regional differences in observed population dynamics. Other
factors being equal, this would most likely reduce the magnitudes of adjustment re-
sponses of economies including a lower level of interregional goods flows and less
pronounced changes in households’ welfare.
Second, the model abstracts from any government activity and in particular does
not consider a public sector pension system. As pointed out above, recent studies
emphasize the role of a public pension system in determining households’ savings
behavior. The model takes an extreme standpoint on this issue by assuming the
presence of a fully-funded pension system which implicitly operates through private
life-cycle savings decisions of households.
Third, in reality there are numerous obstacles preventing international goods mar-
kets from being fully integrated. As long as barriers to trade exert a symmetrical ef-
fect on both regions, the qualitative implications of the analysis should be preserved.
However, cross-country differences in the active set of trade policy instruments are
likely to result in an asymmetric distribution of gains and losses from trade.
Fourth, and probably most critical, is the fact that we carry out the analysis in an
idealized neoclassical world. For instance, introducing elements of imperfect com-
petition, increasing returns to scale and making different assumptions about labor
markets are needed to enhance the policy relevance of the simulation experiments.
Also, it has been shown that if the basic dichotomy of the Heckscher-Ohlin frame-
work is extended to account for multiple skill-related categories of workers (Wood,
1994), country groups (Davis, 1996) and traded goods (Feenstra and Hanson, 1996),
then the main distributive prediction of the Heckscher-Ohlin theory is theoretically
undetermined. Whether the inclusion of those factors would materially alter the con-
clusions is a question for future research.
3.5 Concluding Remarks 73
This appendix provides a sketch of the major issues involved in numerical simula-
tion of the model. In particular, we (i) briefly explain the calibration to a steady
state, (ii) provide an overview of the adopted solution algorithm, and (iii) fully char-
acterize the general equilibrium of the numerical model in a mixed-complementarity
format.
We use the calibration procedure for OLG models which is described in detail in
Rasmussen and Rutherford (2004). The OLG economy is calibrated to an initial
steady state equilibrium in which all quantities grow with constant rate 1 + γ̄ and
present value prices decline with the interest rate 1 + r. The virtue of this approach
is that the calibration of the OLG demand system is independent from the specific
model structure of the production side. The calibration procedure is carried out in
two steps. Demographic shocks are imposed as a counterfactual scenario and are
not incorporated in the initial steady state calibration procedure.
The first step solves for the optimal behavior of a single reference generation
taking into account the existence of aggregate consistency conditions which ensure
that individual choices by OLG households match the aggregate behavior of the
economy. Since in a steady state we know the full path of present value prices, op-
timal behavior for all generations (and household types) can be inferred from the
optimal behavior of a single generation. We pose the household calibration model
as a mixed-complementarity problem (MCP) allowing for corner solutions in labor
supply (we indicate the associated complementary variable to each equilibrium con-
dition using the “perp” operator, “⊥”). Table A–3.3 defines all model variables and
parameters. For generation g = 0, the solution to the household utility maximization
problem given by (3.2) is fully characterized by the following system of equations
(3.23) – (3.35) (since both regions share the same steady state equilibrium, the re-
gion index is dropped):
Definition of full consumption:
νh 1/νh
νh
za,h = αh ca,h + (1 − αh ) `a,h ⊥ za,h (3.23)
∂U(ca,h , `a,h , ρ)
= λh p̄a ⊥ ca,h (3.24)
∂ ca,h
∂U(ca,h , `a,h , ρ)
= ηh ⊥ `a,h (3.25)
∂ `a,h
FOCs for price of labor time:
Aggregate consumption:
ca,h
ccch = ∑ ⊥ ccch (3.32)
a (1 + γ̄)a
Aggregate assets:
ccmah = ∑ cmaa,h ⊥ ccmah (3.33)
a
∑ ccch = C̄ ⊥ ρ (3.34)
h
To fix aggregate assets at the benchmark value select ωscale such that:
To calibrate the model to the benchmark values of aggregate consumption and ag-
gregate assets, we add two additional constraints to the household problem and en-
dogenize the discount factor and a scaling factor on the periodic time endowment.
In the second calibration step, the results from the household calibration model
are used to set up the entire baseline path for all time periods and including those
generations that were born prior to year zero. This involves extrapolating the optimal
behavior of the reference generation (including all different household types) by
making use of the steady-state assumption.
To solve the OLG economy that has been formulated in section 3.2, we use a de-
composition algorithm which has been described in Chapter 2. In what follows, we
provide a brief description of how the algorithm can be applied in the present context
of a two-country, two-sector OLG model. Each iteration of the algorithm comprises
the following steps:
1. Obtain the general equilibrium prices of the “related” Ramsey growth problem.
The “related” Ramsey growth problem is characterized by a model in which the
demand system of OLG households has been replaced by a single representative
infinitely-lived agent. In this model, all other structural elements are identical to
those in the underlying OLG economy.
2. Given general equilibrium prices from the representative agent model, evaluate
OLG households’ demand functions. This step retains the full structure of the
OLG demand system but suppresses any general equilibrium feedback effects.
3. Use the quantity choices from the partial equilibrium relaxation, to recalibrate
the preferences of the representative agent.
4. Return to Step 1 and repeat the procedure until the algorithm converges.
We now formulate the equilibrium of the “related” Ramsey optimal growth model
as a mixed-complementarity problem. The structure of equilibrium is exploited by
the GAMS program and the solver we use to compute the transition and steady state
of the infinite-horizon economy.24 Rutherford (1995b) and Mathiesen (1985) have
shown that a complementary-based approach is convenient, robust, and efficient. A
characteristic of economic models is that they naturally involve a complementary
problem, i.e. given a function F: Rn −→ Rn , find z ∈ Rn such that F(z) ≥ 0, z ≥ 0,
and zT F(z) = 0. A fundamental advantage of formulating an economic model as a
24 For an overview of applied general equilibrium modeling with GAMS (General Algebraic
Modeling System) and the GAMS/MPSGE subsystem see e.g. Rutherford [1998,1999a], and
http://www.mpsge.org.
76 3 Trade Liberalization and Global Demographic Change: A Quantitative Assessment
ν/(ν−1) 1−1/ν
1/(1−ν) ν/(ν−1)
Λr,t pr,c,t + (1 − Λr,t )1/(1−ν) pr,l,t = pr,z,t ⊥ Zr,t (3.41)
∂ pr,a,t
∑ Yi,r,t = ∑ Ar,t ⊥ pi,t (3.43)
r r ∂ pi,t
∂ pr,z,t
∑ Cr,t = Zr,t ∂ pr,c,t (1 + τr,t ) ⊥ pr,c,t (3.44)
r
∂ pi,t
Kr,t (1 − τr,t ) = ∑ Yi,r,t ⊥ pr,re,t (3.46)
i ∂ pr,re,t
∂ pi,t
Lr,t = ∑ Yi,r,t ⊥ pr,l,t (3.47)
i ∂ pr,l,t
income definition:
T
incr = ∑ pr,l,t Ωr,t + pr,k,0 ∑ kr,g,g + T Kr pktr ⊥ incr (3.51)
t=0 g
T denotes the last period of the numerical model. To approximate the underlying infi-
nite horizon economy by a finite-dimensional complementarity problem we choose
a “state variable targetting” approach as proposed by Lau et al. (2002). The infinite
horizon economy can be decomposed into two distinct problems where one runs
from 0, ... , T and the other one runs from T + 1, . . . , ∞.25 Both subproblems are
linked through the post-terminal capital stock in period T + 1. The level of post-
terminal capital is computed endogenously by requiring that investment grows at
the same rate as sectoral output or any other “stable” quantity in the model (see
(3.53)).
The current account premium, τr,t , adjusts to satisfy the additional constraint
(3.52) which ensures that trade is balanced in each period.
The share parameters Λr,t and Θr,t are initially chosen in a way such that the “re-
lated” Ramsey optimal growth problem displays the same baseline steady state path
as the underlying OLG economy. Ωr,t denotes the time endowment of the represen-
tative agent which is a productivity-adjusted sum of the time endowments of OLG
households.
Let p denote the price vector of equilibrium prices that are obtained from solving
the “related” Ramsey optimal growth problem from the previous step. Given the
CES preferences of OLG households, it is possible to obtain from the optimization
problem in (3.2) closed-form demand functions for all generations and all household
types:
25 Note that this method for approximating the infinite horizon relies on the assumption of time-
Variable Definition
ca,h Material consumption by household type h at age a
`a,h Leisure consumption by household type h at age a
za,h Full consumption (composite of leisure and material consumption)
λh Shadow price of lifetime income of household type h
ηh Shadow price of time of household type h
la,h Labor supply by household type h at age a
ωscale Scaling factor on time endowment (parameter used for calibration)
pza,h Price of full consumption for household type h at age a
caa,h Present value of assets over the life cycle for household type h at age a
cmaa,h Value of assets held by age and household type
ccch Total consumption by households of type h
ccmah Total value of assets by households of type h
ρ Discount factor (parameter used for calibration)
Parameter Definition
r̄ Steady state interest rate
γ̄ Steady state population growth rate
p¯a = (1 + r̄)−a Steady state index of present value prices
αh Weight on material consumption for household of type h
σh Intertemporal elasticity of substitution for household of type h
σcl,h Intratemporal elasticity of substitution for household of type h
πa,h Index of labor productivity for household type h at age a
K̄ Benchmark value of capital stock
C̄ Benchmark value of aggregate consumption
x = D (Ψ , p) (3.54)
where x represents a generic variable representing cr,g,h,t and `r,g,h,t , and Ψ denotes
relevant household parameters. Note that evaluating x at p is equivalent to solving a
partial equilibrium relaxation of the underlying OLG economy in which all general
equilibrium interactions due to quantity adjustments are suppressed.
The last step of each iteration involves updating share parameters Λr,t and Θr,t ac-
cording to:
pr,c,t ∑tg=t−N ∑H
h=1 cr,g,h,t
Λr,t = t H
(3.55)
∑g=t−N ∑h=1 pr,c,t cr,g,h,t + pr,l,t `r,g,h,t
∑tg=t−N ∑H
h=1 pr,c,t cr,g,h,t + pr,l,t `r,g,h,t
Θr,t = (3.56)
∑t=0 ∑g=t−N ∑H
T t
h=1 pr,c,t cr,g,h,t + pr,l,t `r,g,h,t
80 3 Trade Liberalization and Global Demographic Change: A Quantitative Assessment
Table 3.4 List of variables and parameters (“related” Ramsey optimal growth problem)
Variable Definition
Ur Utility for representative agent in region r
incr Lifetime income of representative agent in region r
Yi,r,t Output of sector i in region r at time t
Ar,t Final good (CES aggregate of tradable Y ’s) in region r at time t
Zr,t Full consumption (material consumption + leisure) in region r at time t
Ir,t Aggregate investment demand in region r at time t
Kr,t Aggregate capital stock in region r at time t
T Kr Aggregate post-terminal capital stock in region r
Cr,t Aggregate consumption demand in region r at time t
pr,a,t Price of final good at time t in region r
pr,c,t Price of consumption at time t in region r
pi,t Price of tradable sectoral output i at time t
pr,u Price of a unit of utility for representative agent in region r
pr,l,t Market wage rate at time t in region r
pr,z,t Price of full consumption at time t in region r
pr,re,t Capital rental rate at time t in region r
pr,k,t Price of capital at time t in region r
pktr Post-terminal price of capital in region r
τr,t Current account premium at time t in region r
Parameter Definition
Λr,t Weight on material consumption
Θr,t Weight on periodic full consumption in intertemporal utility
αi Capital share parameter in sector i
αa Share parameter in production of A
ν Elasticity parameter (material consumption vs. leisure)
εi Elasticity parameter in sector i production (capital vs. labor)
εa Elasticity parameter in production of A (commodity i vs. j)
δ Periodic capital depreciation rate
Ωr,t Aggregate time endowment at time t in region r
where prices represent general equilibrium prices from Step 1 and quantities repre-
sent partial equilibrium choices obtained under Step 2. Similarly, the time endow-
ment of the representative agent, Ωr,t , is updated in each iteration. Subsequently, the
algorithm involves re-solving the “related” Ramsey optimal growth problem which
is given by (3.36)–(3.53) using the updated values for Λ and Θr,t .
Chapter 4
Quantifying the Sectoral and Distributional
Effects of Demographic Change in Germany
4.1 Introduction
0 ,7 6 0 ,0 0 4
0 ,7 4
w a p r
p o p g ra te 0 ,0 0 2
w o r k in g - a g e p o p u la tio n r a tio
0 ,7 2
p o p u la tio n g r o w th r a te
0 ,0 0 0
0 ,7 0
0 ,6 8 -0 ,0 0 2
0 ,6 6
-0 ,0 0 4
0 ,6 4
-0 ,0 0 6
0 ,6 2
0 ,6 0 -0 ,0 0 8
2 0 1 0 2 0 2 0 2 0 3 0 2 0 4 0 2 0 5 0
Y e a r
Fig. 4.1 Evolution of working-age population ratio and population growth rate
Note: Own calculations based on demographic projections by the Federal Statistical Office of Ger-
many, Eleventh Coordinated Population Forecast 2005, Variant 1-W2. The working-age population
ratio is here defined as the population aged 20-64 as a fraction of total adult population (aged 20-
94). Population growth rates refer to the growth rate of the adult population.
the level of transfer income. Abstracting from this heterogeneity would not allow a
meaningful analysis of the distributional consequences of changes in factor prices.
The results of this chapter are as follows. Over the main projection period from
2003-2050, the demographic transition in Germany will bring about a 7% decrease
in output per capita. This is largely driven by a significant decrease in aggregate la-
bor supply, about 17% over the same horizon (as we model a labor-leisure trade-off
this number already incorporates general equilibrium reactions by households), and
a substantial decline in investment rates (minus 26% in 2050). Population aging is
shown to imply a “capital deepening” process which is reflected by an increase in
the capital-labor ratio of around 2%. As the aging of society shifts the age compo-
sition of the population towards older households that (on average) hold a larger
amount of assets, capital becomes relatively abundant, while labor, also due to the
slower growth in the arrival of new workers, becomes relatively scarce. The higher
capital-labor ratio reduces the need for firms to invest in new physical capital, and
thus investment rates fall. Real factor prices reflect the varying scarcities of factors
of production, and it is projected that the real capital rental rate (real wage rate) falls
(increases) by 0.5% (0.5%) in 2010 and by about 1.3% (1.3%) in 2050.
The demographic transition induces substantial changes in the sectoral composi-
tion of output. Sectoral changes—as measured by a change in the share of sectoral
output in total domestic output—range between about −8.5% and +6.5% in 2010
and between −23.5% and +20.0% in 2050. Expanding sectors are characterized
by higher growth rates of sectoral employment of labor and capital relative to the
economy-wide growth rates of labor and the capital stock. Among the sectors that
benefit most from the demographic transition endowments due to population aging
are Public Services, Textiles and leather products, Education, and Health. Output
shares for the latter two sectors increase by almost one fifth between 2003 and
2050. The sectors that contract most, relatively to total domestic output, are Met-
als, Machinery (around −10%) and Construction, Energy, and Water (−23.51%).
Personal Goods, Mining, and Real Estate Services contract by less (around −2%
to −4%). Output shares for the remaining sectors Mineral and Chemical Products,
Food Products, Printing and Publishing, Wholesale and Retail Trade, Transporta-
tion and Communication, and Banking and Insurance Services change by a rela-
tively small extent (around −1.7% to +1.7%).
Accounting for structural changes in life-cycle consumption that are due to age-
specific preferences does not affect the qualitative results of the benchmark model,
and has only minor quantitative effects for aggregate variables. This underlines that
the nature of the macroeconomic transition is predominantly shaped by the negative
labor supply shock. On the sectoral level, however, demand-side induced effects
stemming from age-dependent consumer spending are found to be quantitatively
important.
In order to evaluate the welfare consequences of the demographic transition we
ask the following question: suppose households of different age and type alive in
2003 were to live through the economic transition with changing factor prices in-
duced by the demographic change. Then how would their welfare change relative
to a situation without a demographic transition ? We find that young households
84 4 Quantifying the Sectoral and Distributional Effects of Demographic Change in Germany
experience substantial welfare gains because they benefit from a future path of in-
creasing wage rates and do not suffer strongly from large losses of capital income on
already accumulated financial wealth. Newborn households in 2003 obtain welfare
gains in the order of 14% and more in lifetime income. In contrast, older asset-rich
households tend to gain less or even lose because of declining interest rates. From
an intra-generational perspective, we find that those members of society for whom
labor income constitutes a smaller part of (future) resources, i.e. households with
relatively low labor productivity, benefit less from the demographic transition. Wel-
fare changes for households with the highest labor productivity are estimated to
be almost twice as large in terms of equivalent variation in lifetime income as for
the least productive households. This underlies the need to account for intra-cohort
heterogeneity.
The demographic transition is projected to increase income inequality in Ger-
many, indicated by an increase in the Gini coefficient by 8.4%. The growing dis-
persion in household income is driven by a rise in capital income inequality and
overcompensates the projected decrease in labor income inequality. As capital in-
come is proportional to wealth, and because the demographic transition induces a
shift towards older asset-rich households, capital income inequality is bound to in-
crease. Labor income inequality decreases because a future path of increasing wages
implies a flatter labor supply profile over the life cycle. We also find that intergen-
erational factors rather than a redistribution between different types of households
within a particular age group account for the observed evolution of income inequal-
ity.
Despite the economic relevance of sectoral changes as an adjustment mechanism
to demographic change, we are only aware of one example in the literature that ad-
dresses this issue. Fougere, Mercenier and Merette (2007) construct a multi-sectoral
OLG model for the Canadian economy which features, among other factors, seg-
mented labor markets and age-dependent preferences. Their analysis focuses on
labor market effects of population aging that are induced by shifts in the sectoral
composition of output and does not investigate the consequences for welfare and
the distribution of income and wealth. Compared to their model, we include en-
dogenous labor supply, intra-cohort heterogeneity, and also allow for international
borrowing and lending, whereas they require trade to be balanced in each period.
More generally, this chapter relates to the literature that uses models and tech-
niques pioneered by Auerbach and Kotlikoff (1987) to study the economic conse-
quences of population aging. The majority of these studies pays special attention to
the impact of aging on the viability of social security systems in closed economies
(see e.g. Huang et al. (1997), De Nardi et al. (1999), Abel (2003), and Fehr et al.
(2004a) ) or in open economy settings ( Attanasio et al. (2006), Börsch-Supan et al.
(2006)). Relative to the literature we see the contribution of this chapter in (i) taking
into account sectoral adjustments that influence the economy’s transition path (ii) in
evaluating the welfare consequences of population aging per se and not just the alter-
native social security reform scenarios and (iii) in incorporating intra-cohort hetero-
geneity which allows to analyze the distributional consequences of changing factor
prices due to the demographic shifts. A similar type of welfare analysis is carried out
4.2 Model Description 85
Hi,t X i,t
σi,HX
Y i,t
σi,Y
VA i,t M i,t
σi,KL σA
of the households side partly draws on data from the 2003 German Income and
Expenditure Survey. Finally, demographic data is taken from the Federal Statistical
Office of Germany and serves as the major exogenous driving process.
4.2.1 Firms
4.2.1.1 Production
Figure 4.2 provides a schematic overview of the nesting structure of sectoral pro-
duction. In each production sector i, j = 1, . . . , I, a representative firm produces a
homogeneous output (Yi,t ). We use a nested constant-elasticity-of-substitution (CES)
production function to reflect empirical evidence on the substitution possibilities. In
the top nest, a material composite (Mi,t ) is combined in fixed proportions with aggre-
gate value added (VAi,t ), i.e. σi,Y = 0, ∀i. M consists of intermediate inputs with fixed
coefficients (Leontief production structure), whereas VAi,t consists of capital (Ki,t )
and labor (Li,t ) services, trading off at a constant elasticity of substitution. Adopting
the calibrated share form (see Appendix 4.6), the unit cost function for each sector i
can be written as2 :
2 There are no explicit production functions in the model because all necessary information is
contained in the dual cost functions.
4.2 Model Description 87
with
" 1−σVA,i 1−σVA,i # 1−σ1VA,i
pl,t pr,t
CVA,i,t = βiL + (1 − βiL ) (4.2)
pl,t pr,t
pA, j,t
CM,i,t = ∑ , i 6= j (4.3)
j χj
where CVA,i,t and CM,i,t =cost index of intermediary aggregate VA and M, respec-
tively, ΩY,i =scale parameter, βiVA = benchmark value share of VA in total cost,
βiL =benchmark value share of L in VA aggregate, pl,t =wage rate, pr,t =rental rate
of capital, pA, j,t =price of Armington good j, χ j =fixed quantity of Armington good
j, σVA,i =elasticity of substitution between capital and labor in the VA nest. Variables
with a “bar” superscript denote benchmark values.
Capital and labor are perfectly mobile across sectors. Each individual firm is as-
sumed to be small in relation to its respective sector and operates on perfectly com-
petitive markets taking goods and factor prices as given, and earning zero profits.
Cost minimization yields the following demand functions for primary factors at the
sectoral level (applying Shepard’s Lemma):
pr,t σVA,i
Ki,t = Yi,t CVA,i,t (4.4)
pr,t
pl,t σVA,i
Li,t = Yi,t CVA,i,t . (4.5)
pl,t
Demand for intermediate input j by sector i is given by:
The law of motion for the aggregate capital stock is given by:
Kt+1 = (1 − δ ) Kt + It (4.7)
where pk,t =purchase price of capital. Investment goods are produced from sectoral
outputs with fixed production coefficients. A unit of the investment good adds to
next period’s capital stock. The investment technology is given by the following
dual cost function: !
I
IA pA,i,t
Ω I ∑ βi ≥ pk,t+1 (4.9)
i=1 pA,i,t
where βiIA =value share of intermediate good i in total investment, and ΩI =scale
parameter.
where βiX =value share of exports, βiH =value share of domestic consumption
in domestic production, p f ,t =price for foreign exchange, pH,i,t =price for Hi,t ,
pY,i,t =price for Yi,t , tY,i =output tax in sector i, σHX = elasticity of transformation.
Following the small open economy assumption, export and import prices in foreign
currency are not affected by the behavior of the domestic economy. In other words,
the small open economy faces infinitely elastic world export demand and world im-
port supply functions.
Similarly to the export side, we adopt the Armington assumption of product het-
erogeneity for the import side. A CES function characterizes the choice between im-
ported and domestically produced varieties of the same good. The associated pricing
equation is given by:
!1−σAR,i 1−σ1AR,i
pH,i,t 1−σAR,i
pA,i,t (1 + tAR,i ) AH p f ,t
= ΩA,i βi
+ βiAM
pA,i,t (1 + t AR,i ) pH,i,t p f ,t
(4.11)
4.2 Model Description 89
where Ai,t =Armington good, ΩA,i =scale parameter, βiAH =value share of domestic
production in domestic consumption, βiAM =value share of imports, pA,i,t =price of
Armington good, tAR,i = tax on Armington goods, σAR,i = elasticity of substitution
of sector i.
The representation of foreign and domestic goods as imperfect substitutes has
the implication that although there is a constant interest rate on the international
bond market, the domestic interest rate may deviate from the world market rate
during a transition period. This happens because building up the domestic capital
stock requires domestic as well as imported inputs. An increase in the capital stock
thereby drives up the relative price of domestic output and the domestic interest rate
until the economy settles in a new steady state with constant relative prices.
From (4.10), supply of sectoral production i to the domestic and export market is:
−σHX
Hi,t Yi,t pY,i,t (1 + tY,i ) pH,i,t
= (4.12)
H i,t Y i,t pY,i,t (1 + t Y,i ) pH,i,t
−σHX
pY,i,t (1 + tY,i ) p f ,t
Xi,t Yi,t
= (4.13)
X i,t Y i,t pY,i,t (1 + t Y,i ) p f ,t
and from (4.11), Armington demand for domestic and imported variety of good i is
given by:
Ai,t pA,i,t (1 + tAR,i ) pH,i,t σAR,i
Hi,t
= (4.14)
H i,t Ai,t pA,i,t (1 + t AR,i ) pH,i,t
Ai,t pA,i,t (1 + tAR,i ) p f ,t σAR,i
IMi,t
= . (4.15)
IMi,t Ai,t pA,i,t (1 + t AR,i ) p f ,t
4.2.3 Households
The household side of the model is disaggregated both within and across generations.
In each period, eight new households corresponding to eight income classes enter
the model. Each household is characterized by overlapping generations and has a
finite and known lifespan in which they engage in market activities. In each period
over the life cycle, households are endowed with units of productive time that they
allocate between labor and leisure. Households are assumed to be forward-looking
individuals that form rational point expectations (perfect foresight). Each household
maximizes lifetime utility by choosing optimal consumption and leisure paths over
the life cycle subject to a lifetime budget and time endowment constraint. There is
no aggregate or household-specific uncertainty.
90 4 Quantifying the Sectoral and Distributional Effects of Demographic Change in Germany
subject to:
νh
1−νh ! 1−νh 1−νh
νh
zg,h,t cg,h,t νh `g,h,t
= αh + (1 − αh ) (4.17a)
zg,h,t cg,h,t
`g,h,t
κh
1−κh ! 1−κh 1−κh
cNF κh
cg,h,t a1,g,h,t κh
g,h,t
= θhF + θhNF (4.17b)
cg,h,t a1,g,h,t cNF
g,h,t
σNF
1−σNF 1−σNF
cNF
g,h,t ai,g,h,t σNF
= ∑ ϕi,h,t (4.17c)
cNF
g,h,t i∈NF ai,g,h,t
g+N
∑ pa,g,h,t cg,h,t ≤ pk,t kg,h,g + p f ,t bg,h,g
t=g
g+N
+ ∑ pl,t πg,h,t (ωg,h,t − `g,h,t ) + p f ,t ζg,h,t (4.17d)
t=g
`g,h,t ≤ ωg,h,t (4.17e)
cg,h,t ≥ 0 , `g,h,t ≥ 0 . (4.17f)
Figure 4.3 visualizes the nesting structure of lifetime utility. Lifetime utility (ug,h )
is additively separable over time, and instantaneous utility (zg,h,t ) enters according
to a constant-intertemporal-elasticity-of-substitution (CIES) function. θg,h,t = share
4.2 Model Description 91
3 Note that in the present setup, the assumption of multi-stage budgeting is innocuous since the
utility function u is weakly separable and the sub-utility functions z and c are homothetic.
4 In reality, preferences may not only be age- but also cohort-dependent. Empirical evidence for
Germany, however, suggests that cohort effects appear to be small (Börsch-Supan, 2004). The
adopted notion of household preferences thus assumes that preferences vary with age but are stable
across generations. For a more detailed discussion see Section 4.5.1.
92 4 Quantifying the Sectoral and Distributional Effects of Demographic Change in Germany
σh
νh
κh
σNF
ture trade and government deficits over the infinite horizon.5 Finally, in each period
of the life cycle, time allocated to leisure consumption cannot exceed the time en-
dowment ωg,h,t (4.17e). Choices for material and leisure consumption are restricted
to be nonnegative (4.17f).
The dual cost functions that correspond to the CES functions (4.16)–(4.17c) in the
household optimization problem are given by:
" 1
g+N 1−σh # 1−σ h
pu,g,h pz,g,h,t
= ∑ θg,h,t (4.18)
pu,g,h t=g pz,g,h,t
5 The initial distribution of domestic and foreign assets across current old generations is selected
such that the economy is on a balanced growth path (for details on the calibration procedure see
Section 4.3).
4.2 Model Description 93
" # 1
pc,g,h,t 1−νh p`,g,h,t 1−νh 1−νh
pz,g,h,t
= αh + (1 − αh ) (4.19)
pz,g,h,t pc,g,h,t p`,g,h,t
1
!1−κh 1−κ
pNF h
1−κh
pc,g,h,t F pA,i,t (1 + τc ) c,g,h,t
= θh + θhNF (4.20)
pc,g,h,t pA,i,t (1 + τ c ) pNF
c,g,h,t
where pu,g,h =price of intertemporal utility, pz,g,h,t =price for zg,h,t , pc,g,h,t =price
for composite consumption good, pNF c,g,h,t =price index for non-food aggregate,
τc =time-invariant tax on consumption, and p`,g,h,t =price for leisure consumption.
Instead of dividing households into groups of workers and retirees, the model al-
lows for optimal endogenous retirement decisions on behalf of the households. The
equilibrium allocation involves a corner solution with respect to labor supply if the
reservation wage exceeds the market wage rate, i.e. if:
A positive amount of labor is supplied if (4.22) holds with equality. For future refer-
ence, labor supply is denoted by lg,h,t . Household demand functions can be written
as:
ug,h pu,g,h pz,g,h,t σh
zg,h,t
= (4.23)
zg,h,t ug,h pu,g,h pz,g,h,t
zg,h,t pz,g,h,t pc,g,h,t νh
cg,h,t
= (4.24)
cg,h,t zg,h,t pz,g,h,t pc,g,h,t
νh
zg,h,t pz,g,h,t p`,g,h,t
`g,h,t
= (4.25)
`g,h,t zg,h,t pz,g,h,t p`,g,h,t
κh
ai,g,h,t cg,h,t pc,g,h,t pA,i,t
= for i = 1, (4.26a)
ai,g,h,t cg,h,t pc,g,h,t pA,i,t
NF
!κh !σNF
ai,g,h,t cg,h,t pc,g,h,t pc,g,h,t pNF
c,g,h,t pA,i,t
= , for i ∈ NF . (4.26b)
ai,g,h,t cg,h,t pc,g,h,t pNF
c,g,h,t pNF
c,g,h,t pA,i,t
4.2.4 Government
The government collects revenue from levying ad-valorem taxes (tY,i ) on sectoral
production (Xi,t and Hi,t ), and ad-valorem taxes (tAR,i ) on Armington production
(Ai,t ). We assume that the government budget is balanced over the infinite horizon,
and that there is a constant tax rate on consumption (τc ) that is determined endoge-
nously such that:
94 4 Quantifying the Sectoral and Distributional Effects of Demographic Change in Germany
∞ ∞
p f ,0 AG,0 + ∑ Φt = ∑ Γt (4.27)
t=0 t=0
where AG,0 is the initial level of assets held by the government, which enters
the model as an exogenous endowment. Total per period tax revenue is given by:
Φt = ∑Ii=1 tY,i (pX,i,t Xi,t + pH,i,t Hi,t ) + tAR,i pA,i,t Ai,t + τc pC,t Ct , where p’s are rele-
vant prices and Ct denotes aggregate consumption. Government expenditure com-
prises government consumption (Gt ) and total transfers to households (Tt ) which
are denominated in the price of foreign exchange: Γt = pG,t Gt + p f ,t Tt . Government
demand combines inputs from the production sectors in fixed proportions:
I
pG,t pA,i,t
= ∑ βiGA (4.28)
pG,t i=1 pA,i,t
and Ii,t and Gi,t denote the investment and government demand for Armington good
i, respectively. The market clearing conditions for primary factors of production are:
I
Kt = ∑ Ki,t (4.31)
i=1
I
Lt = ∑ Li,t (4.32)
i=1
where aggregate labor supply (Lt ) also has to be consistent with individual house-
hold decisions:
t H
Lt = ∑ ∑ lg,h,t πg,h,t . (4.33)
g=t−N h=1
4.2 Model Description 95
Using (4.12) and (4.14), the market clearing condition for sectoral output i destined
for the home market is:
Yi,t pY,i,t (1 + tY,i ) pH,i,t −σHX Ai,t pA,i,t (1 + tAR,i ) pH,i,t σAR,i
= . (4.34)
Y i,t pY,i,t (1 + t Y,i ) pH,i,t Ai,t pA,i,t (1 + t AR,i ) pH,i,t
The shadow price of “time”, or the price for leisure, adjusts to equalize available
time resources and the amount of time that goes into labor supply and leisure activi-
ties:
ωg,h,t = lg,h,t + `g,h,t (4.35)
where lg,h,t is a complementary variable determined by (4.22). The market clearing
condition for government consumption is given by:
Gt = Gt . (4.36)
The market clearing condition for the capital stock is given by:
t H
Kt−1 (1 − δ ) + It−1 + ∑ ∑ kg,h,g = Kt . (4.37)
g=t−N h=1
This equation simply restates (4.7) and also takes into account initial asset holdings
by households alive in year zero that represent claims on the domestic capital stock
(third summand on the left hand side). A similar version holds for the post-terminal
capital stock (see Appendix 4.6, equation (4.51)). Lifetime utility is given by:
incg,h
ug,h = . (4.38)
pu,g,h
Prices for the consumption aggregates zg,h,t and cg,h,t that enter utility are given by
the following market clearing conditions:
zg,h,t = zD
g,h,t (4.39)
cg,h,t = cD
g,h,t (4.40)
where zD D
g,h,t and cg,h,t denote respective demands given by (4.23) and (4.24). Finally,
we fix the value of the trade deficit over the infinite horizon and it may therefore de-
viate from the baseline level in individual periods. Instead of imposing this relation-
ship directly, however, we require balance in all goods and transfers denominated
in foreign exchange. Because of the assumption of perfect international capital mar-
kets the current value price of exports and imports is constant and we therefore only
operate with a single price for foreign exchange determining the level, denoted by
p f . The present value price for foreign currency is given by p f ,t = (1+r)−t p f where
r denotes the world interest rate. The market clearing condition for foreign currency
is:
96 4 Quantifying the Sectoral and Distributional Effects of Demographic Change in Germany
!
t H
∑ p f ,t Dt − Tt + ∑ ∑ ζg,h,t + bg,h,g = ∑ ∑ p f ,t (IMi,t − Xi,t ) (4.41)
t g=t−N h=1 t i
where Dt denotes the periodic government deficit denominated in the price of for-
eign exchange.
4.2.6 Equilibrium
In this section we discuss the specification of model parameters, the data that is
used as model inputs, and aspects of the numerical implementation. In terms of
6 For an overview of applied general equilibrium modeling with GAMS (General Algebraic
Modeling System) and the GAMS/MPSGE subsystem see e.g. Rutherford(1998,1999a), and
http://www.mpsge.org.
4.3 Calibration and Data 97
The way we use the model may be somewhat unconventional in CGE modeling be-
cause we do not perform counterfactual policy analysis but rather use the model as a
projection tool that is fed with demographic data. The thought experiment which un-
derlies the simulations is the following: we take a time-varying demographic struc-
ture as the exogenous driving process that evolves according to demographic projec-
tions by the Federal Statistical Office of Germany. This transition of the population
structure induces a transition path of the economy.
We calibrate the model to the base year 2003 using Input-Output data for Ger-
many and various other data sources.7 For the present application calibration is a
complex task because non-stationarity of the population data implies that the econ-
omy is not on a balanced growth path.8 Our calibration strategy proceeds in two
steps. First, we assume (contra-factually) an initial steady state situation for the
base year 2003, and exploit this assumption to calibrate a number of household pa-
rameters (see Section 4.3.6).9 Then, in order to find the baseline equilibrium path in
the presence of non-stationary population dynamics and to ensure that the base year
data is reproduced endogenously, it is necessary to impose a number of restrictions
on the entire transition model. In our case, technology parameters ΩY,i , ∀i, are cal-
ibrated such that the model replicates sectoral output data in the first model period.
Similarly, ΩA,i , ∀i, and ΩI are solved for endogenously to match base year values
of Armington production and investment, respectively.
To ensure consistency between household behavior and aggregate benchmark
data, it is necessary to solve endogenously for two of the parameters in the individ-
uals’ optimization problem. Rasmussen and Rutherford (2004) present a calibration
7 Our calibration strategy which focuses exclusively on observations of a base year—as for the
OLG context, e.g., in Jensen and Rutherford (2002) and Rasmussen and Rutherford (2004)—
constitutes one possible way of calibrating OLG models to empirical data. This procedure differs
from the approach in Auerbach and Kotlikoff (1987) or Altig, Auerbach, Kotlikoff, Smetters and
Walliser (2001) where parameters are simply chosen so as to generate “reasonable” outcomes by
informally matching long term averages of statistical data. A more systematic approach would be
to perform a so-called “dynamic calibration” by choosing time-varying model parameters to match
observed time series data. This is an ongoing area of research; see, e.g., Ludwig (2005) who sug-
gests to estimate free model parameters by using method of moments methodology that sets to
zero the average discrepancy between actual and predicted (simulated) values. Future work needs
to compare how alternative approaches to model calibration may influence model results.
8 See, e.g., Wendner (1999) for a detailed discussion.
9 We select initial asset holdings k
g,h,g and bg,h,g by assuming (contra-factually) that the economy
is on a balanced growth path, and use the procedure as described in Rasmussen and Rutherford
(2004). The time preference rate ρ is calibrated such that the value of total household consumption
equals aggregate consumption in the base year.
4.3 Calibration and Data 99
procedure that can be used for situations where the economy is assumed to be on
a balanced growth path. It is, however, straightforward to extend their procedure to
non-steady state situations. In the present case, the nature of the decomposition algo-
rithm that we used to compute the transition path of the economy, and that makes it
possible to solve such a large-scale model in the first place, prevents the application
of such a calibration procedure.10 As a consequence, the value of private consump-
tion and the value of total domestic and foreign assets deviate from the benchmark
data.
We employ a social accounting matrix (SAM) for Germany, which is based on Input-
Output (IO) data for the year 2003 issued by the Federal Statistical Office of Ger-
many, and which distinguishes 17 production sectors.11 Table 4.2 lists the names
(and acronyms) of the industrial sector aggregates and describes the subsumed in-
dustries from the original IO data compounded in each sectoral aggregate. This level
of aggregation captures most of the interesting sectors of the German economy that
would be affected by demographic change. In particular, transportation and commu-
nication, banking and insurance services, real estate services, education, and health
services are included as separate sectors.
For aggregation, we include all 71 industries of the German IO table. We adjust
base year investments to obtain a micro-consistent12 SAM which is consistent with
a momentary general equilibrium in the base year. The calibration of technologies
using base year data on prices and quantities is a standard exercise in CGE modeling,
and hence needs no further discussion (see, e.g., Robinson (1991) and Rutherford
(1999a) for a general discussion and Rasmussen and Rutherford (2004, pp.1405) for
an application in an OLG context). Based on the IO data for Germany, we are able
to obtain numerical values for all share parameters (β ’s and χ’s).
10 Since the algorithm decomposes the OLG economy into a representative agent model and a par-
tial equilibrium problem, and involves an iterative procedure between these two subproblems, im-
posing calibration restrictions on the representative agent model does only work if the parameters
used for calibration do not alter demand functions of OLG households. Otherwise, the algorithm’s
search dynamics are profoundly altered because the endogenized parameters become part of the
adjustment process. In this case, we find that the “Sequential Recalibration” algorithm does not
converge, and is therefore not useful in finding the true equilibrium allocation. This also explains
why we have only used technology parameters for calibration of the aggregate economy.
11 Our SAM is based on the Input-Output table 1.1., Issue 18 (2), Federal Statistical Office of
Germany.
12 The SAM incorporates an internally consistent set of relationships. Among the most important
identities in the SAM are zero profits for each production activity, market clearance for all com-
modities and factors, budget balance for households and the government, and the value of imports
equals the value of exports, net of capital flows. In fact, an Arrow-Debreu type applied general
equilibrium model, such as this one, must be based on a SAM; otherwise the data would be incon-
sistent with the underlying behavioral assumptions of the model (which are based on optimizing
behavior of all agents subject to budget constraints and adding up conditions).
100 4 Quantifying the Sectoral and Distributional Effects of Demographic Change in Germany
Note: 1: For reference we give CPA numbers that identify each single industry that is used in
aggregation. CPA = Classification of Products by Activity according to Eurostat. Abbreviations
used: DL = Dienstleistungen.
Note: The different socio-economic groups represent the eight income classes according to the
German Income and Expenditure Survey (EVS) for the year 2003. 1: Population size refers to the
(extrapolated) number of households in million. Labor share denotes the share of the sum of labor
incomes earned by all households in this income class in the economy-wide labor income.
explains the positive correlation with the level of total household income. In the sim-
ulations later on, transfer income per capita is held constant.
Following Auerbach and Kotlikoff (1987), we assume an age-related labor effi-
ciency endowment profile which has the following form:
where λ0 = 0.032, λ1 = 0.003. Not surprisingly, results are sensitive with respect
to the numerical specification of πg,h,t , and we will explore this in Section 4.5. To
match the observed labor share of each income class, we augment their specifica-
tion by proportionally scaling the labor productivity index by a factor ζˆscale,h . It is
assumed that transfers within each household group between individuals of differ-
ent age are distributed equally, i.e. the level of transfers is kept constant over the
life cycle. We set the value share of material consumption (αh ) equal to 0.8 for all h.
Expenditure shares for consumption of Armington good i (ϕi,h,t ) are taken from the
value shares of aggregate consumption as given by the IO table. Hence, in our bench-
mark model, households across groups and age are assumed to have homogenous
preferences and only differ with respect to labor productivity and transfer income.
We will relax this assumption later on to analyze the effect of structural changes in
the final demand composition of old vis-à-vis young people. Homogeneity of pref-
erences in the baseline considerably eases the calibration of the model and, more
importantly, provides a way of disentangling the demand-side induced effects of
population aging.
102 4 Quantifying the Sectoral and Distributional Effects of Demographic Change in Germany
Calibration of the parameters of the utility function requires the integration of em-
pirical estimates for labor supply and consumption demand. We set the intertempo-
ral elasticity of substitution (σh ) to 0.8, ∀h. In the literature values range from 0.5
(Jensen and Rutherford, 2002) to around 1.0 (Krueger and Ludwig, 2007). The elas-
ticity of substitution between consumption and leisure (νh ) is partly determined by
uncompensated labor supply elasticities. Although there is information in the liter-
ature about uncompensated labor supply elasticities, we are not aware of empirical
estimates that take into account variations with age. Due to this lack of informa-
tion, we cannot calibrate νh , and assume a uniform and time-independent elasticity
for all household groups. Estimates typically lie around 0.8 (Altig et al., 2001) and
1.0 (Jensen and Rutherford, 2002). We set ν = 0.8. Following Börsch-Supan et al.
(2006), the consumption share parameter (αh ) is set to 0.6, ∀h.
To calibrate the remaining elasticity parameters we use empirical estimates on
uncompensated price elasticities of consumption demand. Given data on share pa-
rameters (θ F , θ NF , and ϕi ), and given the nesting structure of utility (see Figure
4.3) we can solve recursively, in line with Rutherford (1995b), for the elasticities of
Forecast 2005, Variant 1-W2 issued by the Federal Statistical Office of Germany.
4.3 Calibration and Data 103
where θiL =value share of labor in total production costs of sector i.17 Available
data on uncompensated elasticities from Böhringer et al. (2005) distinguishes be-
tween unskilled and skilled labor. For the homogeneous labor in the model, we take
averages of these estimates and set: ηAGF = −0.33, ηMIN = −0.16, ηTXL = ηPRP =
ηMIC = ηMET = ηFOP = ηMAC = ηPGO = −0.80, ηCEW = −0.48, ηWRT = ηTRC =
−0.19, and ηBIN = ηRES = ηPBS = ηEDU = ηHEA = −0.44.
The following Armington elasticities are taken from Welsch (2008)18 : σAR,AGF =
0.081, σAR,MIN = 0.915, σAR,FOP = 0.841, σAR,TXL = 0.926, σAR,PRP = 0.300, σAR,MIC
= 1.600, σAR,MET = 0.687, σAR,MAC = 2.360, σAR,PGO = 0.184, σAR,CEW = σAR,WRT =
σAR,TRC = σAR,BIN = σAR,RES = σAR,PBS = σAR,EDU = σAR,HEA = 0.871. All sectoral
aggregates are treated as tradable goods. The elasticity of transformation between
domestically supplied and exported goods is uniformly set to σHX = 2.
The depreciation rate of the capital stock is derived using information from the
Federal Statistical Office of Germany on the (gross) value of capital and deprecia-
tion.19 Based on this, we calculate an annual depreciation rate of δ = 0.029.
17 For the relation between compensated and uncompensated elasticities, see, e.g., Hamermesh
(1993).
18 Due to differences in industry classification, we use the country average of .871 for those sectoral
The depreciation rate is calculated as the the ratio of Abschreibungen auf das Anlagevermögen in
jeweiligen Preisen to Kapitalstock.
104 4 Quantifying the Sectoral and Distributional Effects of Demographic Change in Germany
4.3.7 Government
Across all scenarios, we assume that the government maintains fixed expenditure
shares for all sectoral goods according to the baseline path. It is important to stress
that this assumption neglects to take into account the growing political influence of a
larger share of older generations, something which is likely to affect the preferences
of the government. Reported equilibrium outcomes, e.g., with respect to sectoral
changes in public services and health care therefore have to be interpreted as conser-
vative estimates. Furthermore, to provide a reasonable basis for welfare analysis, it
is critical to assume that government transfers per capita in each age and household
group remain fixed. Lastly, we hold government expenditures constant in per capita
terms. Without this assumption the share of government consumption in real GDP
would become unrealistically large.
The model allows us to investigate a wealth of economic effects arising from the on-
going and projected population aging in Germany. The population dynamics induce
a transition path of the economy both in terms of aggregate and sectoral variables
as well as in terms of cross-sectional and intergenerational distributions of income,
wealth and welfare. The analysis of these changes will provide us with answers as
to how the change in the demographic structure of the economy affects key macroe-
conomic and household variables by changing relative goods prices and returns to
capital and labor.
We compute the model equilibrium from 2003 to 2200 and report simulation
results for the main projection period of interest, from 2003 to 2050. We do not re-
port results after 2050 because these hinge crucially on the assumptions concerning
the projection of demographic data beyond 2050. Actually, simulation results suffi-
ciently close to 2050 are already sensitive to demographic projections made beyond
2050 because of the forward-looking nature of the model, and therefore have to be
interpreted with care. The base year 2003 is taken as the main year of reference to
evaluate future changes in the projected time path of variables.
The benchmark model assumes that household preferences do not vary with age.
In Section 4.5, we will modify this assumption in order to assess the quantitative
importance of demand-side induced effects of population aging that results from
structural changes in the composition of old-age consumption.
Table 4.4 reports on the dynamic impact of population aging on key aggregate statis-
tics. The projected percentage changes are presented with respect to the base year
4.4 Results of the Benchmark Model 105
equilibrium solution in 2003. Output per capita is projected to steadily decrease dur-
ing the transition, ending up in 2050 on a 7% lower level than in the year 2003. The
change in output per capita is mainly due to two effects. As the most direct effect, the
decrease in the overall population means that existing resources have to be shared
by less people. However, the ongoing aging of the population also directly reduces
the labor force which, as a factor of production, suppresses output. Aggregate labor
supply (measured in efficiency units) is shown to drop substantially by around 17%.
This second negative effect largely dominates and output per capita declines. The
consumption-output ratio increases by around 1.4% in 2010 and by almost 10% in
2050. This is consistent with the increasing share of older people in the economy that
are dis-savers and hence consume a larger fraction out of their periodic income as
compared to younger households. Consumption per capita, however, stays roughly
constant. The investment-output ratio drastically drops by about 10% in 2010, and
by 26% in 2050. As the population ages and the labor force declines it is optimal
to reduce the capital with which these fewer workers work. The aging of the pop-
ulation brings about a “capital deepening” which is reflected by an increase in the
capital-labor ratio of around 0.6-2% over the projection period. As the aging of so-
ciety shifts the age composition of the population towards older households that (on
average) hold a larger amount of assets, capital becomes relatively abundant, while
labor, also due to the slower growth in the arrival of new workers, becomes scarce.
The higher capital-labor ratio reduces the need for firms to invest in new physical
capital. Thus investment rates fall. The movement of real factor prices is reflected
by the varying relative scarcities of factors of production during the demographic
transition. We observe that the real capital rental rate is projected to fall by 0.5% in
2010 and about 1.3% in 2050. Real wages follow almost exactly the inverse path of
the rate of return to capital.
Table 4.4 Impact of population aging on key aggregate statistics (benchmark model)
Note: Results refer to percentage changes with respect to the base year 2003. Output is defined as
the sum of domestic sectoral production. Real factor prices are deflated by the consumption price
index. Consumption comprises private and government consumption.
106 4 Quantifying the Sectoral and Distributional Effects of Demographic Change in Germany
This section analyzes the impact of population aging at the sectoral level. In the
benchmark model, we do not consider age-specific preferences, and hence the ef-
fects on sectoral output induced by population aging are merely driven by supply
side factors which result from the decline in the effective labor supply and the “cap-
ital deepening” process. It is evident that the demographic transition induces sig-
Note: Figures refer to percentage changes in the sectoral share of output in total domestic output
relative to the base year 2003. An increase in the share of sectoral output from one period to another
means that this particular sector is growing faster / decreasing less rapidly than total domestic
production.
20 As the study by Fougere et al. (2007) is the only point of reference for the type of analysis
presented here, a couples of remarks are in order. It is important to bear in mind that both mod-
els differ along various dimensions. One major difference is that in their model labor supply is
exogenous whereas the present model features a labor-leisure trade off. Given that the evolution of
aggregate labor supply is a key driver of results, this certainly complicates a comparison. Further-
more, the quantitative nature of the demographic process, which is the major exogenous driving
force of the model, the numerical specification of age-specific preferences, and the underlying
Input-Output data are quite different for Germany and Canada. Notwithstanding these considera-
tions, the macroeconomic predictions of both studies are qualitatively similar. To see this, compare
Table 4.4 with Table 9, p. 705, in Fougere et al. (2007). Due to differences in the sectoral aggrega-
4 Quantifying the Sectoral and Distributional Effects of Demographic Change in Germany
Table 4.6 Impact of population aging on sectoral employment and capital use
Share of sectoral employment in aggregate labor (in %) Share of sectoral capital in aggregate capital (in %)
Sector / Year 2005 2010 2015 2020 2030 2040 2050 2005 2010 2015 2020 2030 2040 2050
AGF 0.664 0.678 0.680 0.680 0.673 0.674 0.670 1.663 1.688 1.694 1.696 1.682 1.685 1.679
MIN 0.612 0.611 0.609 0.608 0.606 0.602 0.594 0.185 0.184 0.184 0.184 0.184 0.183 0.181
FOP 1.884 1.925 1.933 1.934 1.907 1.910 1.898 1.268 1.302 1.309 1.314 1.312 1.323 1.322
TXL 1.078 1.129 1.160 1.191 1.216 1.231 1.210 0.474 0.500 0.514 0.530 0.550 0.562 0.555
PRP 2.212 2.225 2.225 2.222 2.203 2.192 2.170 1.802 1.817 1.819 1.821 1.825 1.826 1.816
MIC 5.296 5.294 5.295 5.296 5.292 5.262 5.206 3.301 3.316 3.319 3.332 3.374 3.379 3.361
MET 5.061 4.958 4.937 4.914 4.877 4.762 4.622 2.567 2.530 2.521 2.519 2.534 2.493 2.434
MAC 20.173 19.822 19.836 19.848 19.809 19.362 18.751 6.475 6.433 6.441 6.477 6.588 6.510 6.346
PGO 0.766 0.769 0.768 0.764 0.752 0.745 0.733 0.344 0.347 0.347 0.346 0.346 0.345 0.341
CEW 4.919 4.674 4.524 4.374 4.213 4.029 3.895 4.666 4.414 4.276 4.140 4.001 3.831 3.717
WRT 16.271 16.480 16.504 16.489 16.331 16.300 16.161 8.335 8.404 8.423 8.426 8.373 8.365 8.324
TRC 5.598 5.664 5.677 5.680 5.645 5.638 5.594 6.115 6.159 6.178 6.190 6.172 6.170 6.144
BIN 4.774 4.844 4.855 4.856 4.819 4.822 4.796 3.614 3.649 3.661 3.666 3.649 3.655 3.648
RES 10.915 10.980 10.972 10.947 10.856 10.809 10.708 48.40 48.42 48.43 48.37 48.08 47.88 47.59
PBS 9.790 9.867 9.894 9.954 10.169 10.497 11.001 6.246 6.267 6.289 6.336 6.494 6.709 7.057
EDU 4.824 4.862 4.885 4.940 5.136 5.397 5.803 0.640 0.642 0.646 0.654 0.682 0.718 0.774
HEA 5.163 5.218 5.245 5.303 5.495 5.768 6.188 3.902 3.925 3.949 3.998 4.156 4.367 4.702
Note: The relative expansion/contraction of sectors induced by population aging as indicated by the changes in sectoral output shares (see Table 4.5) can be
explained by looking at the evolution of shares of sectoral employment and capital use. Whether a sector expands/contracts depends on how factors of
production evolve over time relative to aggregate labor and capital. The statistics displayed above indicate which sectors most productively employ abundant
capital and attract scarce labor. An increase in the sectoral share of labor in aggregate labor means that the contraction of labor that is being employed in this
sector is smaller than the overall decline in aggregate labor supply that is due to population aging. An increase in the sectoral share of capital implies that the
quantity of capital used in this sector contracts less than the economy-wide capital stock.
108
4.4 Results of the Benchmark Model 109
tion of industries and the aforementioned quantitative differences in input data, comparing results
at the sectoral level is not instructive.
21 The Gini coefficient is a summary statistic of the Lorenz curve, and is here calculated as the
relative mean difference, i.e. the mean of the difference between every possible pair of observations
(xg,h,t , xg0 ,h0 ,t ) of a generation g and household type h at a given point in time t relative to the size
of the mean:
n + 1 ∑{g,g0 } ∑{h,h0 } |xg,h,t − xg0 ,h0 ,t |
Gt = . (4.45)
n 2n2 µt
Here, n denotes the total number of households alive at t and µt is the arithmetic mean. The Gini
coefficient ranges from a minimum value of zero, when all individuals have equal income (the
numerator in (4.45) then equals zero), to a theoretical maximum of one in an infinite population
in which all income is concentrated in one hand (to see this, divide (4.45) by 2n2 , substitute the
numerator by an expression which depends on the number of non-zero pairs, and apply l’Hôpital’s
rule as n → ∞).
110 4 Quantifying the Sectoral and Distributional Effects of Demographic Change in Germany
0 ,8
0 ,7
G in i c o e ffic ie n t
0 ,6
0 ,5
G in i c o e ffic ie n t fo r to ta l in c o m e
0 ,4 G in i c o e ffic ie n t fo r c a p ita l in c o m e
G in i c o e ffic ie n t fo r la b o r in c o m e
0 ,3
2 0 1 0 2 0 2 0 2 0 3 0 2 0 4 0 2 0 5 0
Y e a r
Fig. 4.4 Evolution of income inequality (all households)
Note: Gini coefficients are calculated on the basis of the entire sample which comprises households
across different types and time.
plays the Gini coefficients that are based on samples of income observations which
pool households across time for a given household type. Inequality of the intergen-
erational capital income distribution increases by around 18% (for h = 1 from 0.53
in 2005 to 0.63 in 2050) and the pattern of the increase is similar to the evolution
of the corresponding Gini coefficient in Figure 4.4 (Gini coefficients for different
household types move in parallel, hence only those for h = 1 and h = 8 are shown).
The reason for the increase in intergenerational capital income inequality—which
ultimately drives the evolution of total income inequality—is that older households
hold more assets than younger households. Since capital income is proportional to
wealth, and as the demographic transition induces a shift towards older, asset-rich
households, capital inequality is bound to increase. Impacts on the intragenerational
distribution of capital income are minor (not shown).
The decline in inequality of the intergenerational labor income distribution
(again, a similar pattern is observed for different household types) stems from the
significant drop in aggregate labor supply which induces a continuous rise in the
real wage rate. Steadily increasing returns to labor diminish the role played by la-
bor productivity over the life cycle in shaping labor supply decisions: households
face a trade-off between supplying labor in middle ages when their productivity is
high and wages are low and supplying labor in older ages when labor productivity
is relatively low but wages have increased. Labor supply over the life cycle still
follows a hump-shaped pattern but the increase in wage rates in the course of the
4.4 Results of the Benchmark Model 111
0 ,8
G in i c o e ffic ie n t fo r c a p ita l in c o m e (h = 1 )
G in i c o e ffic ie n t fo r c a p ita l in c o m e (h = 8 )
0 ,7 G in i c o e ffic ie n t fo r la b o r in c o m e (h = 1 )
G in i c o e ffic ie n t fo r la b o r in c o m e (h = 8 )
0 ,6
G in i c o e ffic ie n t
0 ,5
0 ,4
0 ,3
0 ,2
0 ,1
2 0 1 0 2 0 2 0 2 0 3 0 2 0 4 0 2 0 5 0
Y e a r
Fig. 4.5 Evolution of intergenerational income inequality
Note: Gini coefficients are calculated on the basis of a subset of households which comprises
households of different generations but holds fixed a particular household type
projection period implies a tendency for a “flatter” labor income profile. This effect
explains the significant drop in intergenerational inequality of labor income from
0.28 in 2005 to 0.15 in 2050.
In our model, population aging has a minor impact on the intragenerational in-
equality of labor income over the life cycle. Figure 4.6 plots the evolution of the
Gini coefficient over the lifetime of a generation. Not surprisingly, the dispersion of
labor income within a generation follows the hump-shaped profile of labor produc-
tivity over the life cycle: increasing wages magnify intra-cohort differences in labor
productivity and hence increase labor income inequality.
Overall, however, intergenerational rather than intragenerational factors account
for the evolution of capital and labor income inequality. Since intergenerational eq-
uity decreases the dispersion of the dispersion of the income distribution during the
demographic transition increases.
0 ,5 0 0
G in i c o e ffic ie n t fo r la b o r in c o m e (g = 2 0 0 5 )
0 ,4 9 8 G in i c o e ffic ie n t fo r la b o r in c o m e (g = 2 0 1 0 )
G in i c o e ffic ie n t fo r la b o r in c o m e (g = 2 0 1 5 )
0 ,4 9 6 G in i c o e ffic ie n t fo r la b o r in c o m e (g = 2 0 2 0 )
G in i c o e ffic ie n t fo r la b o r in c o m e (g = 2 0 2 5 )
0 ,4 9 4 G in i c o e ffic ie n t fo r la b o r in c o m e (g = 2 0 3 0 )
G in i c o e ffic ie n t
0 ,4 9 2
0 ,4 9 0
0 ,4 8 8
0 ,4 8 6
0 ,4 8 4
0 ,4 8 2
2 0 1 0 2 0 2 0 2 0 3 0 2 0 4 0 2 0 5 0
Y e a r
Fig. 4.6 Evolution of intragenerational income inequality
Note: Gini coefficients are calculated on the basis of a subset of households which comprises
households of different types but holds fixed a particular generation.
of factor prices affect welfare? How are welfare gains and losses distributed inter-
and intragenerationally?
For welfare comparison we solve each household’s problem under two different
scenarios. Let denote paging the vector of equilibrium prices as documented in the
previous section. Alternatively, consider a “no aging”-scenario where prices are held
constant at their 2003 value, and let pno aging denote this hypothetical price situation.
As the basis of comparison we use the Hicksian equivalent variation (EVg,h ) which
measures the percentage change in lifetime income that would be equivalent to the
price change as implied by the demographic transition in terms of the impact on
lifetime utility. Using the indirect utility function we can write more formally:
vg,h incno aging,g,h (1 + EVg,h ) , pno aging = vg,h incaging,g,h , paging (4.46)
where incno aging,g,h and incaging,g,h denote lifetime income under the respective sce-
nario. Positive numbers of EVg,h indicate that households obtain welfare gains from
the general equilibrium effects of the demographic changes, negative numbers imply
welfare losses.
Table 4.7 documents these numbers for generations that are alive in 2003. We
make several observations. First, newborn agents experience massive welfare gains
from changing factor prices induced by the demographic transition equivalent of
more than 14% in lifetime income. The demographic transition induces a substan-
tial increase in real wage rates and a future path of declining interest rates. The
dominating effect for newborns is the substantial increase in wage rates because
these agents have not yet accumulated financial wealth and thus do not suffer from a
loss of capital income on already accumulated financial wealth, in contrast to older
4.5 Sensitivity Analysis 113
In this section we discuss how the results derived from the benchmark model hinge
on a number of structural model assumptions as well as on the numerical specifica-
tion of key model parameters. More specifically, we want to assess the quantitative
importance of structural changes in old age consumption that are a result of age-
specific preferences. Futhermore, by means of piecemeal sensitivity analysis, we
identify key model parameters that decisively influence the nature and strength of
the economic adjustment.
114 4 Quantifying the Sectoral and Distributional Effects of Demographic Change in Germany
Sectoral good < 25 25-35 35-45 45-55 55-65 65-70 70-80 80+
AGF 0.021 0.021 0.022 0.022 0.021 0.022 0.022 0.021
MIN 0.015 0.015 0.015 0.015 0.015 0.015 0.017 0.019
FOP 0.177 0.171 0.182 0.181 0.172 0.173 0.174 0.169
TLX 0.064 0.062 0.062 0.060 0.056 0.054 0.052 0.046
PRP 0.027 0.028 0.030 0.029 0.030 0.032 0.032 0.029
MIC 0.069 0.075 0.072 0.077 0.079 0.078 0.071 0.067
MET 0.005 0.005 0.005 0.005 0.006 0.006 0.005 0.005
MAC 0.093 0.100 0.095 0.099 0.096 0.092 0.077 0.065
PGO 0.027 0.029 0.030 0.029 0.031 0.032 0.030 0.029
CEW 0.036 0.036 0.036 0.035 0.037 0.038 0.041 0.045
WRT 0.021 0.023 0.022 0.023 0.021 0.019 0.015 0.011
TRC 0.086 0.072 0.062 0.063 0.055 0.050 0.045 0.039
BIN 0.027 0.027 0.027 0.028 0.029 0.028 0.030 0.039
RES 0.240 0.235 0.238 0.232 0.244 0.247 0.270 0.297
PBS 0.061 0.063 0.065 0.063 0.065 0.068 0.069 0.068
EDU 0.015 0.017 0.017 0.011 0.007 0.005 0.004 0.003
HEA 0.015 0.019 0.021 0.027 0.036 0.041 0.046 0.048
Note: Own calculations based on the German Income and Expenditure survey (Einkommens- und
Verbrauchstichprobe) 2003 and a ”z-matrix (Konsumverflechtungstabelle), both Federal Statistical
Office of Germany.
Apart from the most direct effect of population aging, the substantial decline in the
supply of effective labor, a second demand-side induced effect may be important.
Empirical data for Germany suggests that the composition of consumer spending
varies with age.
Table 4.8 displays the age-specific distribution of spending shares across the 17
sectors in the model.22 Most notably, spending shares for Health, Construction, En-
ergy, and Water, and Banking and Insurance Services rise with age. The spending
share on Health goods is three times larger for households aged 80 and older than for
the youngest age group. Spending shares for Education, Transport and Communica-
tion, Food Products, Wholesale and Retail Trade, and Textiles and Leather Products
decline with age. Given the variation of spending shares with age for a number of
22 The numbers in Table 4.8 are based on household data from the German Income and Expenditure
Survey 2003 (Einkommens- und Verbrauchsstichprobe) issued by the Federal Statistical Office of
Germany. The original data lists age-specific consumer spending by consumption categories that,
however, do not match with the sectoral structure of our model. The age-specific distribution of
spending shares across sectoral goods can be recovered using a so-called “Z-matrix” or “bridge-
matrix” that maps consumer spending from the consumption categories to the 71 industries of the
German Input-Output table.
4.5 Sensitivity Analysis 115
sectors, one might therefore expect additional changes in the sectoral composition
of output as the population shifts towards older people.
This view is supported by Lührmann (2005) who estimates age-specific house-
hold demands for a set of eight composite goods using a quadratic almost ideal de-
mand system based on EVS data.23 On the contrary, a study by Schaffnit-Chatterjee
(2007)—which is however based on a crude back-of-the envelope calculation—finds
that the expenditure shares of consumption categories in aggregate expenditures are
not significantly influenced by population aging. The present study is the first at-
tempt in the literature to quantify these effects using a comprehensive general equi-
librium model.
One remark is in order. The measurement of age-specific consumption behavior
is more complicated than Table 4.8 suggests. As these age-specific spending shares
are calculated on the basis of cross-sectional data, they confound age, cohort, and
time effects. Unfortunately, there is no suitable panel data for Germany available
to address this problem. However, looking at repeated cross-sections Börsch-Supan
(2003) finds that these age-consumption profiles did not change much over the past
20-25 years—hence cohort effects appear to be small and probably do not bias Ta-
ble 4.8 to a substantial extent. Assuming that the pattern of age-specific spending
remains much the same in the future, we are able to compute the adjustment path
for an economy where households’ preferences vary with age. As in the benchmark
model, we assume that households of different type have the same preferences.24
Table 4.9 suggests that accounting for age-specific preferences has only minor ef-
fects in terms of the macroeconomic consequences of the demographic transition.
A comparison with the results displayed in Table 4.4 reveals that all key aggre-
gate statistics move in the same direction. This underlines that the nature of the
economic transition is predominantly shaped by the negative labor supply shock.
Also, from a quantitative viewpoint, both models produce very similar results: in
the model with age-specific preferences, output per capita and effective labor sup-
ply decline slightly more, movements in real factor prices are almost identical, and
the investment-output ratio decreases slightly less whereas the consumption-output
ratio is marginally higher relative to the benchmark model.
On the sectoral level, a similar observation applies. Comparing results in Table
4.10 with the respective sectoral impacts shown in Table 4.5 suggests that whether
a particular sector expands or contracts (relative to total domestic output) is solely
determined by the negative labor supply effect: the pattern of changes in output
23 Her results (compare Table 3, p. 29) suggest significant increases in aggregate expenditure shares
for health (6.6%) and energy (5.3%) and a decline for transport (−7.6%) and leisure (−4.4%). A
major shortcoming of her econometric approach, however, is to ignore possible feedback effects
that stem from changes in relative prices.
24 A remark concerns the calibration of technology parameters which has been carried out under the
benchmark household preference structure. As we use the same values for technology parameters
for this “counterfactual”, the model does not endogenously reproduce the base year values. On the
other hand, recalibrating technologies would change the adjustment of sectoral activity in response
to a demographic shock, and would make a comparison between models difficult. For this reason,
we keep all parameter values—except for those relating to age-sensitive preferences—constant.
116 4 Quantifying the Sectoral and Distributional Effects of Demographic Change in Germany
Table 4.9 Impact of population aging on key aggregate statistics (model with age-specific prefer-
ences)
Note: Results refer to percentage changes with respect to the base year 2003. Output is defined as
the sum of domestic sectoral production. Real factor prices are deflated by the consumption price
index. Consumption comprises private and government consumption.
shares across sectors and time that is obtained from the model with age-specific
preferences is roughly identical to the pattern obtained from the benchmark model.
From a quantitative viewpoint, however, demand side induced effects do matter.
Most obvious, quantitative differences between both models are due to a direct ef-
fect that stems from age-related consumer demand. For instance, the age-specific
increase in the spending share of Health as shown in Table 4.8 translates into an
increase in the respective output share as compared to the benchmark model. Direct
demand side effects that are in line with the movement of age-specific spending
shares are also observed for a number of other sectors, e.g., Public Services, Real
Estate, Banking and Insurance Services, Construction, Energy, and Water. Sectoral
activity is, however, also indirectly affected by changes in intermediate demand.
This explains why sectoral output shares for some sectors do not move in the direc-
tion one might expect from looking at spending shares in Table 4.8. For instance,
output shares for Wholesale and Retail Trade increase relative to the benchmark
model although consumer spending shares diminish with age. Table 4.13 displays
input-output data which underlies the sectoral model structure. Looking at column
“WRT” reveals which sectors use Wholesale and retail trade products and services
as an intermediate input in production. Sectors that heavily rely on Wholesale and
Retail Trade products, i.e. FOP, TRC, RES, PBS, are exactly those that exhibit an
increase (a less strong decline) in sectoral output shares relative to the benchmark
model.
Overall, we can summarize that accounting for structural changes in life-cycle
consumption that are due to age-specific preferences does not affect the qualitative
results of the benchmark model. Quantitative effects on the aggregate economy level
are relatively small. The reason for this is that total consumption expenditures by
older households (aged 60-74) only make up 24% of total private consumption in
2003, and do not exceed 30% in 2050. This suggests that the level and the growth of
old age consumption are both not large enough to translate compositional effects in
household consumption into significant economy-wide effects. For this to happen,
4.5 Sensitivity Analysis 117
either a more sizable shift towards older people in the population or a higher level
of old age consumption per se would be required. In contrast, quantitative effects
on the sectoral level are significant. The presence of intermediate demand linkages
implies, however, that for some sectors the pattern of changes in sectoral output
shares is not in line with the movement of age-dependent spending shares.
Table 4.10 Sectoral change due to population aging (model with age-specific preferences)
Note: Figures refer to percentage changes in the sectoral share of output in total domestic output
relative to the base year 2003. An increase in the share of sectoral output from one period to another
means that this particular sector is growing faster (decreasing less rapidly) than total domestic
production.
Table 4.11 shows the effects of changing key model parameters from the base case
parametrization of Section 4.3. For ease of reference, base case results are displayed
in the first row of Table 4.11. For each parameter that is used in the sensitivity analy-
sis, we report results for one value that is below and one value that is above the base
case. Table 4.11 documents that changing parameter values can produce perceivable
quantitative effects. Most importantly, however, it is found that all qualitative model
results are preserved.
Since the negative labor supply shock significantly drives the overall results, pa-
rameters that govern labor supply decisions of households are found to have the
greatest impact. A lower value for the intertemporal elasticity of substitution means
118 4 Quantifying the Sectoral and Distributional Effects of Demographic Change in Germany
that households are less willing to tolerate fluctuations in consumption over their life
cycle. Given decreasing interest rates which make it harder to accumulate wealth
(and earn capital income), a higher degree of consumption smoothing can only be
achieved if labor supply is more elastic. Hence, in light of increasing wage rates
along the demographic transition, lower values for σh imply a weaker contraction in
aggregate labor supply. A lower value of the intratemporal elasticity of substitution
between leisure and consumption makes labor supply less elastic, and hence the de-
crease in aggregate labor supply is more pronounced. Similarly, a lower weight on
material consumption (a larger weight on leisure) in instantaneous utility implies
that labor supply is less sensitive to changes in the wage rate. The extent of the con-
traction in output per capita for different values of these three household parameters
can therefore be explained by movements in aggregate labor supply.
A lower value for the capital depreciation rate stimulates capital accumulation
and somewhat mitigates the decline in output per capita. A lower value for δ
strengthens the role of capital accumulation relative to endogenous variations in
labor supply as an adjustment mechanism to the demographic transition. A larger
decrease in aggregate labor supply is therefore consistent with a smaller reduction
in output per capita. Uniformly lowering the elasticity of substitution between capi-
tal and labor in value-added for all sectors by 30% has only minor effects. A lower
degree of substitutability means that firms are flexible in adjusting to a situation
with changing factor price changes, and hence losses in output per capita are slightly
higher.
In terms of the impact on factor price changes, the decrease in the real return to
capital ranges from −0.29% to −0.71% in 2010 and from −0.23% to −1.22% in
2040 for the scenarios considered. Increases in real wage rates lie between 0.30%
and 0.72% in 2010 and between 0.25% and 1.40% in 2050. The evolution of income
inequality as measured by changes in the Gini coefficient is found not to be very
sensitive: increases from 2010 to 2040 range between 7.3 − 10.1%.
The sensitivity of results with respect to the Armington elasticity of substitution
is found to be low. This underlines the overall robustness of results. Recall that the
assumption of imperfect substitutability between domestic and foreign goods im-
plies that the domestic interest rate can deviate during the transition period from
the exogenous world market interest rate; in the limit of perfect substitutability the
domestic interest rate would be exogenous in all periods. As such, σAR,i indirectly
affects the capital accumulation process. A lower value of σAR,i means that the econ-
omy cannot rely as much on imported goods to build up its capital stock. This neg-
ative effect is buffered by a smaller decline in aggregate labor supply. Both effects
imply that the capital-labor ratio increases by less.
Parameter Output Eff. labor Cap.-labor Real return Real wage Gini coeff. EVb
per capitaa supplya ratioa to capitala ratea income
2010 2040 2010 2040 2010 2040 2010 2040 2010 2040 2010 2040
4.6 Concluding Remarks
Base case parametrizationc -1.30 -5.79 -1.73 -12.15 1.76 2.08 -0.51 -1.03 0.56 1.24 0.582 0.639 14.71
σh = 0.60 -0.15 -1.42 -1.15 -10.97 1.46 1.56 -0.56 -1.23 0.62 1.47 0.584 0.641 14.62
Intertemporal ES
σh = 0.99 -2.12 -8.10 -2.69 -14.03 2.28 2.78 -0.39 -0.50 0.42 0.57 0.580 0.635 12.22
νh = 0.60 -2.29 -7.97 -2.84 -13.70 2.32 2.61 -0.29 -0.23 0.30 0.25 0.618 0.664 13.92
Intratemporal ES
νh = 0.99 -0.94 -4.89 -1.50 -11.97 1.63 1.98 -0.53 -1.15 0.58 1.37 0.570 0.628 14.15
αh = 0.40 -2.53 -8.57 -3.47 -14.49 2.79 2.85 -0.71 -0.65 0.72 0.67 0.587 0.641 11.92
Consum. share param.
αh = 0.80 -0.84 -5.40 -0.85 -11.53 1.11 1.55 -0.37 -1.15 0.44 1.40 0.576 0.633 14.37
δ = 0.025 -0.95 -4.72 -1.96 -12.71 1.82 1.93 -0.49 -0.76 0.53 0.93 0.583 0.641 14.69
Cap. depreciation rate
δ = 0.035 -1.43 -6.07 -1.43 -11.34 1.62 1.90 -0.49 -1.22 0.52 1.36 0.589 0.632 11.60
low -1.40 -6.13 -1.69 -12.12 1.58 1.62 -0.55 -1.03 0.60 1.24 0.581 0.638 13.53
ES in value-addedd
high -1.25 -5.58 -1.80 -12.23 1.96 2.63 -0.48 -1.02 0.53 1.24 0.582 0.639 13.79
low -1.14 -5.12 -1.26 -11.20 1.48 1.55 -0.43 -0.93 0.50 1.25 0.585 0.640 14.37
ES in Armington productione
high -2.29 -7.98 -2.28 -13.04 1.86 2.01 -0.28 -0.34 0.29 0.38 0.578 0.632 11.58
Note: All results are computed using the benchmark model. ES=elasticity of substitution. a : Numbers refer to percentage change relative to 2003. b : Equivalent
variation in lifetime income (in %) for a newborn generation in 2003 of type h = 1. c : σh = νh = 0.8, αh = 0.6, δ = 0.029. For each parameter, the base case
value lies between the two values used in the sensitivity analysis. d : low (high) refers to a situation where the elasticities of substitution in value-added for all
sectors have been reduced (increased) by 30% relative to the base case parametrization. e : low (high) refers to a situation where the elasticities of substitution in
Armington production for all sectors have been reduced (increased) by 50% relative to the base case parametrization.
119
120 4 Quantifying the Sectoral and Distributional Effects of Demographic Change in Germany
While the analysis is carried out for Germany, some of the results may be applicable
to other major industrialized countries facing similar demographic transitions.
The contribution of this chapter is twofold. First, we argue that shifts in the
sectoral structure constitute an important channel of adjustment to population ag-
ing. We document that the demographic transition is bound to induce significant
changes in the sectoral composition of output ranging—as measured by changes in
output shares—from −8.49% to +6.49% in the year 2010 and from −23.51% to
+19.98% in 2050. While accounting for structural changes in life-cycle consump-
tion due to age-specific preferences does not affect the qualitative results of the
benchmark model, and has only minor quantitative effects for aggregate variables, it
is found that sectoral effects are quantitatively important. Second, we analyze the dis-
tributional and welfare consequences that follow from factor price changes induced
by the demographic transition. Newborn households in 2003 experience substantial
welfare gains in the order of 14% and more in lifetime income that arise from in-
creases in wages and declines in returns to capital. In contrast, older households that
have already accumulated assets and low-skilled households for whom labor income
constitutes a smaller part of future resources benefit less and may even lose. Income
inequality in Germany is projected to increase due to the demographic transition, as
measured by a 8.4% increase in the Gini coefficient between 2003 and 2050.
While this chapter discloses a number of important channels of economic adjust-
ment to an aging population, it also abstracts from several issues that can play a role
in shaping the economic transition. First, it is important to bear in mind that we oper-
ate in a “frictionless” environment where all endogenous adjustments are driven by
relative price changes. This can only partly be justified by the long-run character of
the analysis. Second, as we do not take a stand on how social security deals with pop-
ulation aging we implicitly assume the presence of a fully-funded private pension
system that operates through the optimal life cycle savings behavior of households.
In this light, the results can be viewed as a benchmark scenario that abstracts from
any public sector-induced distortions to private savings decisions. Third, by adopt-
ing the small open economy assumption we rule out the possibility that differences
in the extent and timing of demographic processes across countries can shape the
economic transition. It has been demonstrated that the presence of globally unsyn-
chronized aging patterns may induce a substantial amount of international capital
flows (see, e.g., Attanasio et al. (2006) and Börsch-Supan et al. (2006)) or can affect
international trade (see Chapter 3). Fourth, although the assumption of imperfect
substitutability between domestic and foreign goods implies that the domestic inter-
est rate can deviate from the exogenous world interest rate, the small open economy
assumption limits the impact of population aging on the interest rate. For exam-
ple, employing multi-region OLG models Börsch-Supan et al. (2006) and Krueger
and Ludwig (2007) find that demographic change tends to reduce the interest rate
by about 0.5-1 percentage points whereas the present model projects a decline of
about 0.2 percentage points over a comparable time horizon. Finally, as suggested
by Ludwig, Schelkle and Vogel (2007), higher returns to labor may make it optimal
for young households to obtain a better education, thereby increasing the supply of
4.6 Concluding Remarks 121
effective labor. Thus, accounting for endogenous human capital accumulation may
mitigate the macroeconomic impact of demographic change.
and demands for the individual factors of production (applying Shepard’s Lemma):
Y cy px,i σ
Xi
= (4.49)
X i Y cy px,i
In order to solve a finite approximation of the model with a T -period model horizon
(where T is the last period of the numerical model), we use a state variable targeting
approach as suggested by Lau et al. (2002). The post-terminal capital stock (KT)
can be determined as part of the equilibrium calculation by targeting the associated
control variable investment (IT ) in the terminal period. In the present model this is
achieved by adding the following constraint to the model:
122 4 Quantifying the Sectoral and Distributional Effects of Demographic Change in Germany
IT YAGF,T
= ⊥ KT (4.50)
IT −1 YAGF,T −1
where we use the “⊥” operator to indicate the complementarity aspect. The price
for post-terminal capital (pkt) is determined by a last period version of (4.37):
t H
KT (1 − δ ) + IT = + ∑ ∑ ktg,h,g KT ⊥ pkt (4.51)
g=t−N h=1
where ktg,h,g =assets holdings of terminal generations, i.e. those households that live
beyond T .
Following Rasmussen and Rutherford (2004), government income over the infi-
nite horizon is given by:
1+r 1
gov = p f ,t Tt + ∑ Φt + p f ,0 B − p f ,T B (4.52)
t r r
where the second summand on the right hand side denotes the present value of
future government deficits over the infinite horizon and the third summand gives
the value of terminal assets in the terminal period using the steady-state assumption
to sum over the infinite horizon. Thus, the government agent cannot consume all
of its wealth in the last model period but is “endowed” with a negative amount
of assets that is sufficient to finance future government deficits over the infinite
horizon consistent with a steady state equilibrium at time T . When the government
budget is balanced over the infinite horizon, the terminal level of government assets
is endogenous. This is necessary to account for the fact that a policy change may
cause the government’s asset position in period T to deviate from the baseline level.
Again, following Rasmussen and Rutherford (2004, p.1401), we therefore break
(4.27) into two equations that are imposed on the model as additional constraints:
T
p f ,0 AG,0 + p f ,T AG,T = ∑ (Φt − Γt ) (4.53)
t=0
1
p f ,T AG,T =(ΦT − ΓT ) . (4.54)
r
Equation (4.53) states the required balance within the model horizon as a function
of the endogenous level of terminal assets, AG,T . Equation (4.54) gives the value of
terminal assets as a function of income and expenditure in the terminal period using
the steady-state assumption to sum over the infinite horizon.
4.6 Concluding Remarks 123
TECHNOLOGY PARAMETERS
βiVA benchmark value share of value-added in total costs
βiL benchmark value share of labor in value-added
χi fixed-coefficient (Leontief) in Armington production
βiIA benchmark value share of intermediate good in investment
βiX benchmark value share of exports
βiH benchmark value share of domestic consumption in domestic production
βiAH benchmark value share of domestic production in domestic consumption
βiAM benchmark value share of imports
tAR,i tax on Armington goods
tY,i output tax in sector sector i
σVA,i elasticity between capital and labor in value-added
σHX elasticity of transformation
σAR,i elasticity of substitution in Armington production
δ annual capital depreciation rate
ΩY,i total factor productivity in production of Yi,t
ΩA,i total factor productivity in production of Ai,t
ΩI total factor productivity in composite investment good
HOUSEHOLD PARAMETERS
r annual steady-state interest rate
θhF benchmark value share of food products
θhNF benchmark value share of non-food products
ϕi,h,t benchmark value share of a single non-food product in non-food aggregate
σh intertemporal elasticity of substitution
νh intratemporal elasticity of substitution between consumption and leisure
κh elasticity of substitution between food and non-food products
σNF elasticity of substitution between non-food products
αh benchmark value share of composite consumption
ρ time preference parameter
πg,h,t labor productivity index over the life cycle
4 Quantifying the Sectoral and Distributional Effects of Demographic Change in Germany
is clearly a relevant dimension to the analysis. A public pension system can strongly
influence private savings behavior. Consequently, a remarkable number of papers
investigates the adjustments required in the pension system due to demographic
shifts. Important examples include Huang et al. (1997), De Nardi et al. (1999), and
Abel (2003). Krueger and Ludwig (2007) show that the impact on factor prices
induced by the demographic transition is to some extent influenced by the design
and reform of the pension system. Against this background, the results presented in
this book ought to be viewed as a benchmark world in which there is a fully-funded
pension system in place that operates through private life-cycle savings.
Second, the analyses are carried out in a “frictionless” environment in which all
endogenous adjustments are driven by relative price changes. For instance, the as-
sumption of frictionless labor markets may be particularly critical for the European
economies in which unions play an important role in the wage setting process. Also,
the hypothesis of perfect competition on goods markets may not reflect actual mar-
ket structures for particular sectors of the economy. In general, however, accounting
for market frictions is surely more of an issue for an analysis that is—unlike the mod-
els presented in this book—concerned with the short-run effects of the demographic
transition.
Third, it is assumed that demographic dynamics are exogenous to economic de-
velopment. There is a consensus in the literature that in the long run neither fertility
nor mortality are exogenous to economic growth (see, e.g., Barro and Becker (1988),
Barro and Becker (1989), and Boldrin and Jones (2002)). Furthermore, it is plausi-
ble to assume that international migration is endogenous and reacts to international
differences in income.
Fourth, and maybe most critical is the assumption that individuals are fully ratio-
nal and forward-looking. All models presented here assume that household behavior
follows the rational expectations paradigm. While this assumption considerably re-
duces the complexity of the analysis and, from a practical modeling point of view,
enables a relatively simple implementation of household behavior, there are cer-
tainly good reasons to question the rationality of agents. For instance, empirical evi-
dence (for an overview see Attanasio (1999)) shows that households do not smooth
consumption as much as the life cycle theory suggests. At a more general level, the
economic literature on bounded rationality indicates that households sometimes vi-
olate their preferences (Kahneman, 2003). It is an open question of research how
to integrate such behavioral assumptions into a general equilibrium framework. Pos-
sible routes to achieve this are, e.g., to impose additional constraints on household
behavior while maintaining the assumption of full rationality (Altig et al. 2001) or to
depart from this hypothesis by assuming myopic rule-of-thumb behavior (Campbell
and Mankiw, 1991).
Fifth, as suggested by Ludwig et al. (2007), higher returns to labor may make
it optimal for young households to obtain a better education, thereby increasing
the effective supply of labor. Thus, endogenous human capital accumulation may
mitigate the macroeconomic impact of demographic change.
Sixth, the analyses presented here abstract from intergenerational altruism whereby
successive generations are linked through recursive altruistic preferences that give
5 Concluding Remarks 127
Abel, Andrew B. (2003) ‘The effects of a baby boom on stock prices and capital
accumulation in the presence of social security.’ Econometrica 71, 551–578
Abel, Andrew B., Gregory N. Mankiw, Lawrence H. Summers, and Richard J. Zeck-
hauser (1989) ‘Assessing dynamic inefficiency: theory and evidence.’ Review of
Economic Studies 56, 1–20
Allais, Maurice (1947) Economie et Intérêt (Imprimerie National, Paris)
Altig, David, Alan J. Auerbach, Laurence J. Kotlikoff, Kent A. Smetters, and Jan
Walliser (2001) ‘Simulating fundamental tax reform in the United States.’ Ameri-
can Economic Review 91, 574–595
Altonji, Josef G., Fumio Hayashi, and Laurence. J. Kotlikoff (1992) ‘Is the extended
family altruistically linked? Direct tests using micro data.’ American Economic
Review 82(5), 1177–1198
Ambler, Steve, Emanuela Cardia, and Christian Zimmermann (2002) ‘International
transmission of the business cycle in a multi-sector model.’ European Economic
Review 46, 273–300
Ando, Albert, and Franco Modigliani (1963) ‘The “life cycle” hypothesis of saving:
aggregate implications and tests.’ American Economic Review 53, 55–84
Arntz, Melanie, Stefanie Boeters, Nicole Gurtzgen, and Stefanie Schubert (2006)
‘Analysing welfare reform in a microsimulation-age model: the value of disaggre-
gation.’ Discussion Paper No. 06-076, Centre for European Economic Research
Attanasio, Orazio (1999) ‘Consumption.’ In Handbook of Macroeconomics, ed. J. B.
Taylor and M. Woodford, vol. 1b (Amsterdam: Elsevier Science B. V.) chapter 11,
pp. 741–812
Attanasio, Orazio P., and Giovanni L. Violante (2005) ‘The demographic transition
in closed and open economy: a tale of two regions.’ IADB Working Paper No.
412
Attanasio, Orazio P., Sagiri Kitao, and Giovanni L. Violante (2006) ‘Quantifiying
the effects of the demographic transition in developing economies.’ Advances in
Macroeconomics. Article 2
130 References
Auerbach, Alan, and Laurence Kotlikoff (1987) Dynamic Fiscal Policy (Cambridge
University Press)
Auerbach, Alan, Laurence J. Kotlikoff, and Jonathan Skinner (1983) ‘The efficiency
gains from dynamic tax reform.’ International Economic Review 24(2), 81–100
Backus, David K., Partick J. Kehoe, and Finn E. Kydland (1992) ‘International real
business cycles.’ Journal of Political Economy 100(4), 745–775
Bajona, Claustre, and Timothy J. Kehoe (2006) ‘Demographics in dynamic
Heckscher-Ohlin models: overlapping generations versus infinitely lived con-
sumers.’ NBER Working paper w12566
Barbie, Martin, Marcus Hagedorn, and Ashok Kaul (2004) ‘Assessing aggregate
tests of efficiency for dynamic economies.’ Topics in Macroeconomics. Article
16
Barro, Richard J., and Gary S. Becker (1988) ‘A reformulation of the economic
theory of fertility.’ Quarterly Journal of Economics 103, 1–25
(1989) ‘Fertility choice in a model of economic growth.’ Econometrica 57, 481–
501
Barro, Robert J. (1974) ‘Are government bonds net wealth?’ Journal of Political
Economy 82, 1095–1117
Barro, Robert J., and Xavier Sala-i-Martin (2004) Economic growth, 2nd ed. (The
MIT Press)
Batini, Nicoletta, Tim Callen, and Warwick McKibbin (2006) ‘The global impact of
demographic change.’ IMF Research Paper
Bloom, David E., and Jeffrey G. Williamson (1998) ‘Demographic transitions and
economic miracles in emerging Asia.’ World Bank Economic Review 12(3), 419–
455
Boeters, Stefan, Christoph Böhringer, Thies Büttner, and Margit Kraus (2008) ‘Eco-
nomic effects of VAT reform in Germany.’ Applied Economics, forthcoming
Böhringer, Christoph, and Thomas F. Rutherford (2007) ‘Innovation, uncertainty
and instrument choice for climate policy.’ unpublished manuscript
Böhringer, Christoph, Stefan Boeters, and Michael Feil (2005) ‘Taxation and unem-
ployment: an applied general equilibrium approach.’ Economic Modeling 22, 81–
108
Boldrin, Michele, and Larry E. Jones (2002) ‘Mortality, fertility and saving in a
malthusian economy.’ Review of Economic Dynamics 5(4), 775–818
Börsch-Supan, Axel (2003) ‘Labor market effects of population aging.’ Labour
17, 5–44. Special Issue
(2004) ‘Global aging - issues, answers, more questions.’ MEA Discussion Paper
55-04, MEA, University of Mannheim
Börsch-Supan, Axel, Alexander Ludwig, and Joachim Winter (2006) ‘Aging, pen-
sion reform, and capital flows: a multi-country simulation model.’ Economica
73, 625–658
Bourguignon, Francois, and Amedeo Spadaro (2006) ‘Microsimulation as a tool for
evaluating redistribution policies.’ Journal of Economic Inequality 4(1), 77–106
Bourguignon, Francois, Anne-Sophie Robilliard, and Sherman Robinson (2005)
‘Representative versus real households in the macroeconomic modeling of in-
References 131
Fehr, Hans (2000) ‘Pension reform during the demographic transition.’ Scandina-
vian Journal of Economics 102(3), 419–443
Fehr, Hans, Gitte Halder, and Sabine Jokisch (2004a) ‘A simulation model for the
demographic transition in Germany.’ Working Paper, Universität Würzburg
Fehr, Hans, Sabine Jokisch, and Laurence Kotlikoff (2003) ‘The developed world’s
demographic transition—the roles of capital flows, immigration, and policy.’
NBER Working Paper no. 10096
(2004b) ‘The role of immigration in dealing with the developed world’s demo-
graphic transition.’ NBER Working Paper no. 10512
Feroli, Michael (2003) ‘Capital flows among the G-7 nations: a demographic per-
spective.’ Working Paper, Federal Reserve Bank
Fougere, Maxim, Jean Mercenier, and Marcel Merette (2007) ‘A sectoral and occu-
pational analysis of population aging in canada using a dynamic cge overlapping
generations model.’ Economic Modelling 24, 690–711
Frankel, Jeffrey, and David Romer (1999) ‘Does trade cause growth?’ American
Economic Review 89(3), 379–399
Galor, Odor (1992) ‘A two sector overlapping generation generations model: a
global charaterization of the dynamical system.’ Econometrica 60, 351–386
Glenn Harrison, Thomas F. Rutherford, and David G. Tarr (1997) ‘Quantifying the
Uruguay round.’ Economic Journal 107, 1405–1430
Hamermesh, Daniel S. (1993) Labor demand (Princeton University Press, New Jer-
sey)
Harrison, A., and G. Hansen (1999) ‘Who gains from trade reform: some remaining
puzzles.’ Journal of Development Economics 59(1), 125–154
Huang, He, Selahattin Imrohoroglu, and Thomas Sargent (1997) ‘Two computations
to fund social security.’ Macroeconomic Dynamics 1, 7–44
Huggett, M., and G. Ventura (1999) ‘On the distributional effects of social security
reform.’ Review of Economic Dynamics 2, 498–531
Imrohoroglu, Ayse, Selahattin Imrohoroglu, and Douglas H. Joines (2003) ‘Time
inconsistent preferences and social security.’ Quarterly Journal of Economics
118(2), 745–784
INGENUE (2001) ‘Macroeconomic consequences of pension reforms in Europe:
an investigation with the INGENUE world model.’ Working Paper, available at:
http://www.cepii.fr/anglaisgraph/workpap/pdf/2001/wp01-17.pdf
Jensen, Svend E. Hougaard, and Thomas F. Rutherford (2002) ‘Distributional ef-
fects of fiscal consolidation.’ Scandinavian Journal of Economics 104(3), 471–
493
Jones, Ronald (1965) ‘The structure of simple general equilibrium models.’ Journal
of Political Economy 73(6), 557–572
Kahneman, Daniel (2003) ‘Maps of bounded rationality: psychology for behavioral
economics.’ American Economic Review 93(5), 1449–1475
Kehoe, Timothy J., and David K. Levine (1985) ‘Comparative statics and perfect
foresight in infinite horizon economies.’ Econometrica 53(2), 433–453
Koopmans, Tjalling C. (1965) ‘On the concept of optimal economic growth.’ The
Econometric Approach to Development Planning
References 133
Kotlikoff, Laurence (2000) ‘The A-K OLG model: its past present and future.’ In
Using dynamic general equilibrium models for policy analysis, ed. W.H. Glenn,
S.E.H. Jensen, L.H. Pedersen, and T.F. Rutherford (North-Holland, Amsterdam)
Kotlikoff, Laurence J., Kent A. Smetters, and Jan Walliser (1999) ‘Privatizing so-
cial security in the United States: comparing the options.’ Review of Economic
Dynamics 2(3), 532–574
Kotlikoff, Laurence J., Kent Smetters, and Jan Walliser (2001) ‘Finding a way out
of America’s demographic dilemma.’ NBER Working Paper No. 8258
Krueger, Dirk, and Alexander Ludwig (2007) ‘On the consequences of demographic
change for rates of return to capital, and the distribution of wealth and welfare.’
Journal of Monetary Economics 54(1), 49–87
Krueger, Dirk, and Felix Kubler (2006) ‘Pareto-improving social security reform
when financial markets are incomplete!?’ American Economic Review 96(2), 737–
755
Krusell, Per, and Anthony Smith (1998) ‘Income and wealth heterogeneity in the
macroeconomy.’ Journal of Political Economy 106, 867–896
Larch, Martin (1993) ‘Dynamically inefficient equilibria in the Auerbach-Kotlikoff
model.’ Empirical Economics 18, 159–172
Lau, Morten I., Andreas Pahlke, and Thomas F. Rutherford (2002) ‘Approximating
infinite-horizon models in a complementarity format: a primer in dynamic general
equilibrium analysis.’ Journal of Economic Dynamics and Control 26, 577–609
Ludwig, Alexander (2005) ‘Moment estimation in Auerbach-Kotlikoff models: how
well do they match the data?’ MEA Discussion Paper 93-05, MEA, University of
Mannheim
Ludwig, Alexander, Thomas Schelkle, and Edgar Vogel (2007) ‘Demographic
change, human capital, and endogenous growth.’ Working paper, available at:
http://www.mea.uni-mannheim.de/alexludwig/downloads/HK 27Nov07.pdf
Lührmann, Melanie (2005) ‘Population aging and the demand for goods and ser-
vices.’ University of Mannheim, MEA discussion paper no 05095
Mathiesen, Lars (1985) ‘Computation of economic equilibria by a sequence of linear
complementarity problems.’ Mathematical Programming Study 23, 144–162
Meeraus, Alexander (1983) ‘An algebraic approach to modeling.’ Journal of Eco-
nomic Dynamics and Control 5(1), 81–108
Modigliani, Franco, and Richard Brumberg (1954) ‘Utility analysis and aggregate
consumption functions: an interpretation of cross-section data.’ In Post Keyne-
sian Economics, ed. Kenneth Kurihara (New Brunswick, NJ: Rutgers University
Press)
Mountford, Andrew (1998) ‘Trade, convergence and overtaking.’ Journal of Inter-
national Economics 46, 167–182
Negishi, Takashi (1960) ‘Welfare economics and existence of an equilibrium for a
competitive economy.’ Metroeconomica 12, 92–97
Phelps, Edward S. (1961) ‘The golden rule of accumulation: a fable for growthmen.’
American Economic Review 51, 638–641
134 References
Storesletten, Kjetil, Christopher I. Telmer, and Amir Yaron (2007) ‘Asset pricing
with idiosyncratic risk and overlapping generations.’ Review of Economic Dynam-
ics 10, 519–548
Townsend, Robert M. (2002) ‘Safety nets and financial institutions in the
Asian crisis: the allocation of within-country risk.’ International Mon-
etary Fund. Prepared for the IMF Conference on Macroeconomic Poli-
cies and Poverty Reduction, March 14-15, Washington, D.C., available at:
http://cier.uchicago.edu/papers/papers.htm
Townsend, Robert M., and Kenichi Ueda (2003) ‘Financial deepening, inequal-
ity, and growth: a model-based quantitative evaluation.’ IMF Working Pa-
per 03-193. International Monetary Fund, Washington, D.C., available at:
www.imf.org/external/pubs/ft/wp/2003/wp03193.pdf
United Nations (2006) ‘World population prospects: the 2006 revision population
database.’ Population Division, 2006
Ventura, Gustavo (1999) ‘Flat tax reform: a quantitative exploration.’ Journal of
Economic Dynamics and Control 23, 1424–1458
Weil, David (2006) ‘Population aging.’ NBER Working Paper No. W12147
Welsch, Heinz (2008) ‘Armington elasticities for energy policy modeling: evidence
from four european countries.’ Energy Economics 30(5), 2252–2264
Wendner, Ronald (1999) ‘A calibration procedure of dynamic CGE models for non-
steady state situations using GEMPACK.’ Computational Economics 13, 265–
287
Wong, Kar-yiu (1990) ‘Factor intensity reversal in a multi-factor, two-good econ-
omy.’ Journal of Economic Theory 51(2), 434–442
Wood, Adrian (1994) North-South trade, employment, and inequality: changing for-
tunes in a skill-driven world (Clarendon Press, Oxford)